One of the more peculiar things about my life is how completely the texture of my days shifts depending on which city I’m in.

I moved to the UAE in the aftermath of October 20, 2020, with no real plan of returning to live in Nigeria. Life, as it often does, had other ideas. I eventually settled, got married, and now split my time between Lagos and Dubai.

For many reasons, good and bad, I am constantly reminded of the gap between both worlds. Chief among them: when I’m in Dubai, utilities are practically invisible to me. Internet, electricity, water. They exist the way oxygen does. I rarely think about them except when the monthly bill arrives. Something breaks, there’s a number to call.

My life in Lagos is different. Here, utilities are as unpredictable and mystifying as the weather.

Some days everything works and the day hums along fine. Most other days, the power cuts out, or the WiFi drops, or the water pump fails — sometimes all three, without warning.

One minute I’m on a work call triaging a strange edge case, the next, the internet drops and I’m scrambling to reroute the call through my phone. Or I’m keeping one eye on the generator’s fuel level, hoping I don’t need to run two streets over to the filling station in the middle of the workday. And if I’m especially unlucky, a neighbour messages to say my water tank is leaking and asks if I want the number of a good plumber.

None of this feels as dramatic as it sounds when you’re living through it. It doesn’t even qualify as distraction anymore. It’s just a layer of dysfunction tacked onto my daily life — “infrastructure management” if you may, that I run quietly in the background if I want anything like peace.

In one city, I start work and forget that infrastructure exists. In the other, I spend energy and slices of attention managing it. A single difference that reshapes almost everything about how productive my day can be.

It took me an embarrassingly long time to connect that experience to the work I do at Learning Equality.

I’ve spent the last few years working with the team at Learning Equality, building and improving a platform called Kolibri.

Kolibri is an offline-first learning platform built for communities where internet access is exactly as I’ve described: intermittent, expensive, or simply absent. It’s present in over 220 countries and territories, at refugee camps, rural schools, and correctional facilities. Places where a reliable connection is a luxury rather than infrastructure.

When I first joined, I understood this intellectually. One third of the world lacks reliable internet access. Seven in ten children in low income countries struggle to read and understand simple text. The numbers are stark. I could recite them in my sleep within weeks of joining.

But there’s a gap — almost as wide as my life in Lagos versus Dubai — between knowing a statistic and understanding what it actually means for a person trying to teach, trying to learn, trying to make something of their day.

Most software assumes the Dubai version of the world. Reliable internet. Instant responses. Continuous connectivity. Content streams from the cloud. Progress syncs automatically. Updates arrive quietly in the background.

Take that network away and the whole model starts to wobble.

Kolibri was designed to assume the opposite. Instead of assuming the internet will always be present, it assumes the opposite. Content lives locally. Entire libraries sit on a small server inside a school. Teachers and students access everything over a local network that keeps working even when the outside internet disappears.

But the architecture is only half the story. The human side is where things get interesting.

In some schools, teachers download new content overnight because that’s when the connection is fastest. The students arrive in the morning and the lessons are ready. Other times a sync fails halfway through. The connection drops before the download finishes. The update that was supposed to arrive before class simply doesn’t.

And then you see the rapid improvisation that happens in classrooms everywhere.

A teacher reshuffles the lesson plan. A student switches to a different exercise. An admin plugs in a USB drive carrying content from another school down the road.

Watching this happen changes how you think about products. When connectivity is unreliable, the most important feature isn’t speed or elegance. It’s whether the system keeps working when the network disappears. Whether a user can recover when something fails halfway. Whether the product lets people continue instead of forcing them to start again. These are questions many designers still treat as purely technical — but are in truth really about understanding and solving for the environments users actually live in.

Learning Equality was built around this reality from the start. The mission isn’t to just distribute content — it’s to make the highest quality educational tools and resources reachable in places where the usual assumptions about infrastructure don’t hold.

That has led to some unusual deployments. Kolibri runs on small local servers where the outside connection appears only a few hours each week. Learning libraries have been carried across regions on hard drives before ever syncing with a wider network. The platform is designed so that a fully configured library can be prepared anywhere and deployed in environments with little or no connectivity — with educators able to start using it immediately.

One conversation that has stayed with me was with Cynthia, an educator who prepared a full Kolibri library from her base in the US and supported deployments across schools in Enugu and Lagos. What struck me was how unremarkable the process sounded. The content could be moved, copied, and used without needing a live connection or continuous coordination.

The system wasn’t built to depend on who shows up. It didn’t assume perfect infrastructure. It simply worked within the constraints. And the educators on the ground could take it from there.

When the internet behaves like weather, the product has to behave more like a shelter than a pipeline.

It has to keep working when conditions shift.

Living between Lagos and Dubai has made this impossible to ignore.

In one place, infrastructure fades into the background and you design assuming the network is always there. On the other hand, connectivity is something people manage day by day, plan around, work despite.

The second world is where most of the world actually lives. And designing for it; really designing for it, not just bolting on an offline mode as an afterthought; means starting from a different question entirely.

Not: what happens when the internet goes out?

But: what happens when it only comes on briefly?

Somewhere, in a classroom far from the data centres where most software is built, a teacher downloaded lessons last night on a borrowed connection. This morning her students arrived. The network is gone. The lessons are still there.

That gap — between when the connection existed and when the learning happened — is exactly what good design is supposed to hold open.

— — — — —

Further reading:
- October 20, 2020: Lekki toll gate shooting
- Smart school deployments: CNI Summer Series · UB Spotlights · Cynthia Tysick on LinkedIn
- Kolibri: About Kolibri
- Learning Equality: 2025 in Review · Support the Mission here

Support Learning Equality’s work:

• Donate: No amount is too small. Every little bit can help us ensure underserved learners and educators are receiving the educational opportunities they deserve.
• Engage with us on LinkedIn, Instagram, YouTube, Tik Tok, and Facebook.
• Stay in the loop by subscribing for quarterly updates.

When the Internet Behaves Like Weather was originally published in Learning Equality on Medium, where people are continuing the conversation by highlighting and responding to this story.

What does learning look like when it speaks without sound?

On November 20, 2025, I drove along the GT (Grand Trunk) Road, one of Asia’s oldest highways, from Faisalabad to Sadhoke, to visit a satellite school run by FESF (Family Educational Services Foundation), an organization dedicated to providing quality education and empowerment for Deaf children in Pakistan through its Deaf Reach Program.

Turning off the GT Road onto an unpaved bumpy path where the GPS stopped trying, I wasn’t sure what we would find. It was my first time visiting a school in Pakistan where Kolibri, Learning Equality’s offline-first learning platform, was being used with Deaf learners. In a country where education is chronically under-resourced, and where disability deepens inequity, each functioning school feels like a small miracle. Being a little lost, we asked for help from a passerby who quickly guided us to the school.

Stepping into the Sadhoke classroom felt like a trip back to my own school days in Pakistan. The hum of energy, the rhythm of lessons, small rituals I remembered from childhood, all were there. But alongside the familiar, something new was quietly at work: Kolibri, supporting teachers and students in real time, even when resources were limited.

According to Deaf Reach, there are over one million Deaf children in Pakistan, yet fewer than 5% have access to formal education. Deaf Reach Schools (DRS) is the only nationwide network reaching both rural and urban communities. Their approach is holistic: literacy and skills training, teacher and interpreter development, Pakistan Sign Language programs for parents and communities, advocacy, job placement, and digital content development. Today, DRS has 38 locations — 9 main campuses, 9 satellite schools, and 20 digital classrooms, serving over 2,000 students, with a PSL dictionary of more than 6,000 words, helping expand and standardize sign language across Pakistan.

Stepping into the classroom, the first thing I noticed was the absence of spoken words, and the flurry of intentional hands. Saying “Assalamualaikum” to the teacher, I waved to the students, who immediately stood to greet me, curiosity lighting their faces. I found myself mentally shifting from English to Urdu, stitching together thoughts that felt suddenly too slow. And then, as if to bridge any communication gap, a small bouquet of roses appeared in my hands, one student had made it. A simple gesture, showing that connection doesn’t always need sound.

I settled into a seat at the back as the lesson began. The topic of the day was plants. Haseeb Nasrullah, the teacher, used Kolibri projected on an LED screen to guide the class, then invited students to the whiteboard to label each part of the plant. Midway through, we were reminded that this was still a small village in Pakistan, the electricity suddenly cut out, plunging the room into darkness. But Haseeb and her colleague, Mehak Fatima, didn’t pause for even a breath. Power outages are routine here, and while the solar backup took its time to kick in, they shifted seamlessly to workbooks and kept the learning moving.

The satellite school is modest, just two classrooms and about fifteen students ranging from early childhood to their mid-teens, yet nothing about it feels small. The teachers rely on content developed in-house by FESF’s Digital Learning Program Team, who script lessons, film videos, and polish instructional materials with care. The tech team keeps everything running smoothly so the learning experience stays seamless. Parents who were once unsure about sending their children now trust the school deeply; many even ask for sign language classes so they can communicate more confidently at home. Everything here signals thoughtfulness, adaptability, and a genuine understanding of what these students need to thrive.

One of Learning Equality’s core values is making sure that storytelling flows both ways. We come to learn — really learn — how Kolibri lives in these classrooms: the wins, the challenges, the realities. We also create space for teachers, administrators, and learners to ask us questions, and to shape the conversation. Beyond the predictable curiosities (including whether I was married), many wanted to understand my role and why I had come. When the teacher signed “K” for Kolibri, the students immediately lit up. Watching them break into the Kolibri-Fly, hands fluttering like small wings, was a moment that pulled everything into focus. After more than a decade of work, these are the moments that remind us exactly why we do what we do.

After an inspiring morning in Sadhoke, I made my way to FESF’s main campus in Lahore to meet their regional team. The moment I stepped in, the scale and intentionality were unmistakable. Bright classrooms sit alongside a vocational training center and industry partnerships, KFC, Sapphire, and others, that create real employment pathways for Deaf adults. As I walked through the campus, it was clear that every detail had been carefully considered. Each space, each tool, each program reflects an underlying philosophy: that Deaf learners deserve an ecosystem built around their success, not one that asks them to adapt to scarcity.

Leaving Sadhoke and Lahore, I kept returning to a single question:

If learning is ultimately about connection, how can technology widen the road of opportunity, especially for learners with disabilities who sit at the sharpest edges of the digital divide?

Watching teachers and teams work with such care reminded me that meaningful education has never been about curriculum or software alone. It begins with centering learners, honoring their agency, and meeting them where they are. When that alignment happens, possibility opens wide.

For Learning Equality, every opportunity to be welcomed into a classroom, in all its diversity, is a profound privilege. We are proud to support FESF in creating accessible, offline-first learning opportunities that allow Deaf students in Pakistan to dream big and pursue extraordinary futures. When technology, pedagogy, and human connection reinforce one another, the learning environment becomes expansive, a place where every gesture, sign, or smile counts as much as a spoken word.

If this story moved you, let it move you into action.

Explore Kolibri. Learn a few words in Pakistan Sign Language. Share FESF’s Deaf Reach work. Tell someone why Deaf education matters, and why inclusive edtech must be part of the solution. Donate and support Learning Equality. Together, we move closer to a future where learning is accessible to all.

Where Learning Happens Without Sound: A Visit to Deaf Reach Schools was originally published in Learning Equality on Medium, where people are continuing the conversation by highlighting and responding to this story.

This blog post was authored by Anna French, Learning Equality’s Digital Storytelling and Communications Manager.

By the time Mary said goodbye, I could barely see her face.

She is the head teacher at YIDA in Uganda, a refugee-led nonprofit. We had started our video interview at six o’clock in the evening her time, nine in the morning mine. Me at a desk in rural Colorado, her in the early childhood development center where she’d walked to after her school day ended, because the connection there was slightly more reliable than at the primary school. A generator hummed somewhere behind her voice. A single classroom light held back the dark for a while.

Somewhere in the hour between her six and her seven, it stopped being enough. Her voice stayed warm and clear. But the light had gone, and I was left speaking to a silhouette, a darkened rectangle where a person had been. She was telling me about the barriers to providing quality education in a refugee settlement, and the small ways technology had been helping chip away at them. The conversation kept going. The screen held what it could.

I have spent time in rural East Africa before: a month in a Ugandan village with a local nonprofit, and after that, six months living and working at a children’s foundation on the shores of Lake Victoria in Tanzania. I know something of that light, how fast it goes once the sun sets away from the city. But I was here, in Colorado, watching the rectangle go dark. Familiarity is not presence, and I felt the difference.

I work as a Digital Storytelling and Communications Manager at Learning Equality (LE), a nonprofit that created Kolibri, a digital education platform designed for communities where reliable internet, electricity, and resources cannot be assumed. My job is to tell the story of what that technology makes possible. The way I do it is entirely through a screen. In my three years working remotely with LE, I have never visited the specific classrooms, prisons, refugee settlements, and villages where our work lives in the field. What I have done is spend hours on video calls, listening.

Mazin Mukhtar runs Akshar Foundation in India, an organization that has schools built specifically for children living in poverty, many of them former child laborers pulled from stone quarries and wood-cutting factories, boys as young as ten, and girls pulled from household work. He told me about the model they developed to get them to stay: “learn and earn.” Let the older kids tutor the younger ones and explain to the parents that the more a child learns, the more they earn. He said it plainly, with the ease of someone who has turned a hard idea over for years until he found the clean edge of it.

“We use child labor to end child labor.”

I wrote it down. There are sentences that rearrange something in you. His was one of them.

Kolibri, he told me, solved something his team struggled with. Because Akshar doesn’t sort children by age, each child moves through math and language at their own pace, and tracking every student’s progress had become a “logistical nightmare.” When they found Kolibri, each child got their own account, their own coursework, their own learning path. Digital skills acquired organically, “baked into the cake” as he put it.

Mercy Nantongo at Village Enterprise in Uganda told me about a business mentor turned digital scaling officer named William. William trains first-time entrepreneurs in rural Uganda, people starting savings groups and learning to run small businesses, many holding a digital device for the very first time. Before they adopted their SPRINT program using Kolibri, a mentor like William could only be in one village at a time. Now, two mentors can run the same lesson simultaneously across sixty villages. William’s cohort became the first in Uganda’s program history to cross a 93% graduation rate above the poverty line.

Mercy also told me about a breastfeeding mother who would miss parts of sessions. Because the lesson lived on a tablet within the community, operated by a trained neighbor, she could come back and replay exactly what she’d missed. As a former breastfeeding mother myself, that detail has stayed with me more than almost any statistic I’ve gathered. It is so ordinary. So specific. So quietly human.

Then there is Habiba. Habiba Ayman is a computer engineering student in Egypt. She spent her summer as a Google Summer of Code contributor to Learning Equality, building a new text editor for Kolibri Studio, the tool educators use to write and format learning content. She is a native Arabic speaker, and she made sure the editor worked just as naturally for right-to-left languages as for left-to-right ones. She thought about who would be sitting down to write a lesson in Arabic, and she built for that person.

She said something in our interview that I keep returning to. Talking about the accessibility decisions she made, the choices about how to balance performance with inclusion, she put it simply:

“As developers, the choices we make decide who gets included and who gets excluded.”

Habiba wasn’t speaking abstractly. She was speaking from experience, as someone who has navigated systems not always built with her in mind, building something so that door stays open a little wider for the next person.

I think often about the parallel between what I do and what Learning Equality does. Kolibri exists because quality education has always been gated by geography, infrastructure, cultural and gender norms. Technology, built with humility about what it can and cannot do, becomes a bridge across that gap.

My video calls are their own low-budget version of the same logic. Learning Equality doesn’t have the resources to send documentary crews to East Africa or South Asia. Most small nonprofits don’t. But a video call to capture a human story costs almost nothing. And what gets lost in translation, the heat, the weight of a place, the particular quality of light at six in the afternoon in Uganda, is not always what matters most. Sometimes what matters most is that someone on the other side of the screen trusts you enough to say something real.

Even the tools I use to conduct these interviews reflect that same philosophy. I record on Riverside, mostly because it records to each person’s local device rather than relying on a single internet stream. That single design decision means that when someone’s connection drops in northern Uganda or rural Bangladesh, the recording doesn’t. Their words don’t disappear into a bad signal and the story is protected.

There are AI tools baked into that workflow too, features that help clean audio, and generate transcripts and captions after interviews are captured. I use them, because they save time. But I want to be honest about what they are and what they aren’t: they are the back-end of a process whose entire value lives at the front end, in the moment someone decides to tell you something about their world. No model trained on the internet could have predicted what Mazin would say, the specific gentleness in the way Mercy described a breastfeeding mother, or the thoughtfulness behind Habiba’s smile. AI helps me polish the recording. It does not help me know what to listen for.

It’s a philosophy Learning Equality carries into its own AI work. The tools being built around Kolibri are designed to support educators and curriculum specialists, not replace them. A human in the loop, not as a safeguard but as the point. The technology surfaces what’s possible. The person decides what matters.

This is more important now than it might have been a few years ago, because we are living through a technological moment that is genuinely strange. What a small team can generate and distribute today would have been unimaginable a decade ago, and we are all still catching up to what that means for how we work, what we value, and who gets displaced. And yet amid that acceleration, the gap between who has access to these tools and who doesn’t is widening at the same pace. What hasn’t changed is who gets access first, and who waits.

The risk I see most clearly, working in this space, is not only that AI will replace communicators, though I do see that pressure building. It’s that it will make it easier to skip conversations entirely. To summarize instead of listen. To generate a story about people in low-resource communities rather than ask them about their own lives. The efficiency gains are real. So is what gets lost in them.

The people I interview are navigating the same tension from the other side. Kolibri exists because someone asked: what does education look like when you build it around the infrastructure people actually have, rather than the infrastructure they might have soon? The platform works offline because connectivity could not be assumed, not tomorrow, not even in ten years. These were not AI-generated insights. They were the result of people staying in close enough relationship with the communities they were building for.

That is the work. In communications, in education technology, in any field that claims to serve people it doesn’t always see directly: to stay accountable to the people at the other end of the work, because they are the ones who know what they need, and given the chance, they will tell you. The tools change. That doesn’t.

Mary is the story. William is the story. The breastfeeding mother who got to replay a lesson she missed. Mazin, who found that the problem contained its own solution. Habiba, who thought about who would be writing in Arabic and built for her.

I couldn’t be in the room with any of them. But I have spent time in places like some of theirs, in villages in Uganda and Tanzania, in the particular darkness that comes when the sun sets completely. I have seen enough to know what it means when someone builds a door and leaves it open.

I witness most of my sunrises these days out the window of a small farm in Colorado. The light here is different. But the hour between Mary’s six and seven still finds me, and I am still opening the screen the way you open a window, hoping something real gets through.

What the Screen Can’t Hold Alone: was originally published in Learning Equality on Medium, where people are continuing the conversation by highlighting and responding to this story.

Since 2018, Learning Equality has worked to simplify curriculum alignment — a critical but complex process for educators. While Open Educational Resources (OERs) abound, organizing them to local standards remains a persistent challenge, especially in low-resource contexts where time, personnel, and tools are scarce. Our early efforts focused on digitizing curriculum standards into machine-readable formats, laying the groundwork for efficient discovery and adoption. More recently, we’ve taken a major step forward with the release of AI powered recommendations in Kolibri Studio, our online curricular tool for curating OERs for use offline in the Kolibri Learning Platform

This feature marks Phase 1 of our roadmap toward semi-automated curriculum alignment. Future phases will introduce curriculum parsing (Phase 2) and end-to-end channel generation (Phase 3), combining recommendations with structured curriculum data to create fully aligned learning pathways. But first, let’s explore the technical foundation powering these recommendations.

The Problem

Kolibri Studio bridges the gap between open content and local learning needs, enabling educators to curate materials from the Kolibri Library — a 700GB+ repository spanning over 173.000 resources in 120 languages, organized into “channels” or custom collections. Yet the Library’s scale presents a paradox: abundance complicates discovery.

For educators and curriculum specialists, aligning resources to local standards has traditionally been a manual, time-intensive process. The effort required scales with channel size: From hours for small collections to weeks for comprehensive curricula. In rural schools, refugee camps, and other under-resourced settings, this burden diverts limited time from teaching, planning, and learner support. While the Kolibri Library offers unparalleled breadth, its depth demands better tools to surface relevant content efficiently.

The Approach

Our AI powered recommendations feature directly addresses the core challenges of OER curation by reducing manual effort through context-aware suggestions, preserving educator agency with human-in-the-loop validation, and adopting an iterative development approach that incorporates real-world usage patterns. This balanced strategy ensures we deliver immediate value while building a foundation for continuous improvement based on actual educator needs and behaviors.

The project centers on three technical pillars:

1. Curriculum Automation API: A dedicated service to generate and serve recommendations.
2. Studio Integration: Seamless highlighting of suggestions into the curation interface (covered in Part 2).
3. Feedback Loop: Systems to capture implicit and explicit user interactions (also detailed in Part 2).

The Technical Implementation: From Competition to Production

Kaggle: helping us lay the foundation

Our technical journey began not with architecture diagrams, but with a fundamental challenge: Could machine learning automate the manual, resource-intensive process of curriculum alignment? To answer this, we partnered with The Learning Agency Lab and UNHCR to host a Kaggle machine learning competition focused on developing models that could predict how well educational content matched specific curriculum topics.

The competition surfaced a critical insight: an ensemble of specialized embedding models could effectively capture the nuanced relationships between diverse educational materials and curriculum standards. Using a dataset of 20 million+ content items and 10,000+ educator-curated folders spanning multiple languages and subjects (with particular emphasis on STEM), participants developed solutions that significantly outperformed manual alignment processes. The winning ensemble became the scientific foundation for our recommendation system.

Transcription of multimedia

In moving toward a more comprehensive semi-automated curriculum alignment tool, we also identified the need for audio and video transcription. While a significant portion of the Kolibri Library’s audio and video materials already include captions, there still exist some without. In order to thoroughly align those multimedia resources, we needed a thorough human-in-the-loop transcription feature. During the 2023 Google Summer of Code (GSoC), we mentored a contributor who prototyped a transcription editor into Kolibri Studio using OpenAI’s Whisper. This GSoC project has provided us the necessary UI features to eventually integrate a full preprocessing pipeline for all multimedia OER materials.

From Research to Production

With a proven ensemble of models in hand, we faced our next challenge: building a production system capable of serving these recommendations at scale while maintaining the precision required for educational contexts. This meant addressing several key architectural dimensions:

1. Data Representation: Structuring Content for Semantic Matching

The competition demonstrated that semantic embeddings could effectively measure curriculum alignment, but we needed a robust pipeline to generate and serve these embeddings in real-time. We designed a standardized text processing approach that preserves both content and context.

For OER Content:

For Destination Folders:

This structured format ensures consistent embedding generation while maintaining the hierarchical relationships that educators rely on when organizing curriculum materials.

2. Database Architecture: Balancing Performance and Practicality

Given our team’s familiarity with PostgreSQL, we initially considered pgvector for vector storage. However, to accelerate development and avoid database migration overhead during this critical phase, we opted for Google Cloud Firestore with its native vector support. This decision enabled us to:

• Deploy quickly using existing cloud infrastructure
• Leverage serverless scaling to handle variable workloads
• Maintain flexibility for future migration to pgvector as our needs evolve

Each embedding is stored with a content hash, enabling efficient updates without redundant computations — a critical optimization given our 700GB+ library and the constant influx of new materials.

Distributed Pipelines with Ray

The ensemble’s computational demands and the scale of our content library required a distributed architecture. We implemented Ray, which provided the ideal balance of flexibility and performance, to power two distinct but complementary pipelines:

Embed Pipeline

1. Preprocessing: Extracts and cleans text from diverse OER formats (EPUBs, PDFs, HTML5, exercises, videos)
2. Embedding Generation: Passes preprocessed text through our model ensemble
3. Storage Action: Stores generated embeddings in Cloud Firestore with metadata hashes

Recommendations Pipeline

1. Context Preprocessing: Creates a breadcrumb path from the current folder structure in Kolibri Studio
2. Context Embedding: Generates embeddings for the folder context using the same ensemble
3. Recommendation Action: Performs parallel KNN searches across all model embeddings and combines results using weighted averaging

Both pipelines follow the same fundamental pattern of preprocessing → embedding → action, but operate on different inputs (content vs. context) and produce different outputs (stored embeddings vs. ranked recommendations). The parallel processing capabilities of Ray enable simultaneous queries across all models in our ensemble, while the hash-based storage system ensures we only recompute embeddings when source content changes.

Building for Evolution: Versioned Architecture

From the outset, we designed our architecture to evolve alongside our understanding of curriculum alignment. The pipeline’s modular construction — comprising distinct preprocessors, embedding models, and action handlers — naturally lent itself to versioning. Rather than treating this as an afterthought, we embraced it as a core capability, leveraging Ray’s distributed execution model to create what we call “presets”: complete, versioned configurations of our recommendation system.

This approach transforms the recommendation engine into a set of distributed building blocks that can be:

• Composed flexibly: Mix and match components to test new approaches
• Deployed independently: Update individual elements without system-wide changes
• Scaled selectively: Allocate resources precisely where needed

The result is an architecture that thrives on our continuous improvement upon it.

Orchestration: The Role of Agents

While our versioned, distributed building blocks provided the necessary flexibility, they introduced a new challenge: managing the complexity of multiple versions running concurrently. This is where our Agents come into play. Not to be confused with AI agents, these agents are orchestration components that serve as the control layer for our recommendation system.

Agents handle the operational complexity that arises from our modular architecture by:

• Managing component lifecycles: Creating, reusing, and shutting down Ray Actors (workers) as needed for each preset
• Providing stable interfaces: Exposing consistent API contracts through proxies while internal implementations evolve
• Routing intelligently: Directing requests to appropriate versions based on configuration
• Ensuring resilience: Handling component failures and automatic recovery

This orchestration layer allows our pipeline code to focus solely on recommendation logic, while Agents handle the distributed execution details. The result is a system that can:

• Deploy new API versions cleanly (e.g., /v2 with improved models)
• Scale expensive components (like GPU workers) independently
• Test new approaches without disrupting existing users
• Roll back instantly if issues arise

The Path to Offline

While the current system runs in Kolibri Studio, we’re actively working toward distilling the ensemble into a single, smaller model for offline use in the Kolibri Learning Platform. This will require significant real-world usage data, gathered from feedback on this recommendations system, to ensure the distilled model maintains the ensemble’s alignment quality.

Liked what you read? Stay tuned for more!

In Part 2 of this series, we’ll cover:

• Studio Integration: How recommendations appear in the UI.
• The Feedback Loop: Implicit and explicit signals that refine the engine.
• Pilot Impact: Early data on the usage of this feature.

Get Involved

Educators: Try the recommendations feature in Kolibri Studio and share your feedback.

Developers: Stay tuned for when we open source the Curriculum Automation API.

Researchers: Help us improve alignment for underrepresented languages and subjects.

You can also support our work by:

• Donating: No amount is too small. Every little bit can help us ensure underserved learners and educators are receiving the educational opportunities they deserve.
• Engaging with us on LinkedIn, Instagram, YouTube, Tik Tok, and Facebook.
• Staying in the loop by subscribing for monthly updates.

AI Powered Recommendations in Kolibri Studio: Looking Under the Hood was originally published in Learning Equality on Medium, where people are continuing the conversation by highlighting and responding to this story.

Over the last number of years, we’ve collaborated with institutions, partners, researchers, community members, and the private sector to explore ways that AI could support teaching and learning by helping support tasks that are time-intensive. Let’s talk about how we’re doing it.

We know that AI can transform work that previously took months to something that can be finished in a few weeks, days or even hours. To optimize time-consuming, repetitive tasks, AI can be a powerful tool. One of the most time consuming tasks? Curriculum alignment.

Curriculum alignment is the process of organizing, adapting, and contextualizing resources to the standards and learning objectives of the national curriculum or textbook that is relevant to the learners and educators being served by a program — and is a critical ingredient in enabling discovery and use of these materials to support effective learning.

This release builds on years of foundational work by Learning Equality, Vodafone Foundation and UNHCR to improve the process of bringing together relevant, supplemental content that can support teaching and learning.

It is an example of equitable AI use that focuses on teacher support, not replacement. By giving teachers their valuable time back, they can focus on what matters most: supporting learners.

What’s new:

All Kolibri Studio users will have access to recommendations when importing content from the Kolibri Library. While you can still use our general keyword search and browse favorite Channels in the main part of the page, on the side, you’ll now see a new view of recommended content.

The recommendations are based on the information you add to your folders, including folder title, description and language. These details are crucial to help Kolibri Studio find content tailored to the topics you’re choosing materials for. So the more details you add to the description field or folder title to indicate level or focus area, such as “Primary 4” or “Advanced Biology” the better / more relevant your recommendation results will be!

Here’s how to use the new features:

Key takeaways:

1. Specific folder names help! Since the recommendation includes information with your folder title, as well as the parent folders, setting up a folder structure that contains relevant, specific information will help you get the best results.

2. Import content on a per-folder basis for more specific results quickly. To save time, we recommend structuring your project first, especially if you’re working from a curriculum document or syllabus. Navigate into each folder separately, and you will see more context-specific recommendations for that specific topic or unit of materials.

3. Engaging with the feature improves it over time. We collect data from interactions — including whether you add a resource to your channel, ignore it, or say that it’s not relevant. Taking a few moments to give us this feedback will help us refine the feature and provide you with better recommendations in the future.

Why this is particularly important to us…

Since its inception, Learning Equality has focused on making high-quality Open Educational Resources available offline, making sure they support the needs of national curricula. In low-resource contexts, this alignment is essential: it ensures that teachers and learners can discover and engage with content that is both relevant and accessible.

Learning Equality’s AI work is the result of cross-sector collaboration, with contributions from a number of key individuals and organizations who have worked with us for years, helping guide our thinking and processes. This includes early backing from Google.org for our collaboration with UNHCR and Vodafone Foundation to explore semi-automation of curriculum alignment; the educators in Kakuma refugee camp and the dozens of individuals who participated in our design sprints and hackathons; the team at Kaggle, who hosted the machine learning competition that helped develop the algorithms with support from Schmidt Futures, UNHCR, and The Learning Agency Lab, as well as participants and prize winners; the Endless Network and Patrick J. McGovern Foundation, whose support enabled the integration of search recommendations into Kolibri Studio.

Content recommendations in Kolibri Studio are just one component of how Learning Equality has been enabling learners to access relevant content through AI-driven processes.

Tell us what you think!

We couldn’t be more excited to put this work out into the world! We welcome all feedback about your experiences using our AI-generated recommendations feature. So drop us a note on socials or the Community Forum and let us know your thoughts!

— — — — — — — — — — — — — — — — — — — — — — — — — — — — — —

You can also support our work by:

• Donating: No amount is too small. Every little bit can help us ensure underserved learners and educators are receiving the educational opportunities they deserve.
• Engaging with us on LinkedIn, Instagram, YouTube, Tik Tok, and Facebook.
• Staying in the loop with our work by subscribing for monthly updates.

Smarter Content Discovery: AI Powered Recommendations Now in Kolibri Studio was originally published in Learning Equality on Medium, where people are continuing the conversation by highlighting and responding to this story.

AI is entering classrooms faster than our ability to govern it, study it, or fully understand its consequences.

From elite, AI-driven private schools promising hyper-personalized learning, to chatbots quietly reshaping how students study, the narrative is often one of inevitability and scale. At the same time, much of the AI narrative assumes constant connectivity, strong governance, and well-resourced systems. Perhaps most dangerous of all, it also assumes that all children come equally equipped to succeed under minimally guided approaches to learning. They do not.

Real-world examples echo this: A Rest of the World report shows that students leaning on AI tools have struggled academically, and governments have already paused or rolled back AI-led education initiatives after errors and unintended harms surfaced.

Research from the Brookings Institution has found that while overuse can weaken students’ social development, strain relationships with teachers, and introduce serious risks around safety and accuracy, thoughtfully designed AI tools can have the opposite effect, and be used to enrich learning when contextualized and applied as part of a pedagogically sound approach.

That is why, at Learning Equality, we approach AI from an equity and needs-driven lens. We leverage the technology in learning tools that are specifically designed with and for the communities we serve. All our AI work is grounded in meeting real underserved classroom needs, and developed to support teachers in myriad ways: From providing real-time coaching, to helping them find the most relevant resources for a given lesson, to generating formative assessments, our focus is on helping them save time, so they can continue to play the fundamental role of teaching.

In underserved communities, where infrastructure is uneven, teachers are central and learning is deeply social. From this vantage point, equity doesn’t come from replacing educators or automating childhood. It comes from creating opportunities for AI literacy over AI dependence, from access to tools that support teachers rather than sideline them, and from integrating technology into public systems instead of bypassing them.

For funders, policymakers, and technologists, this is a pivotal moment. The question isn’t whether AI has a role in education, but how far, how fast, and under whose terms. If innovation prioritizes scale over care, and speed over evidence, we risk widening the very gaps we claim to close.

As AI reshapes learning, are we building systems that strengthen human connection, or quietly training the next generation to learn alone? The future of education should be shaped with humility, humanity, and those closest to the classroom at the center. And that’s how we’re approaching AI at LE.

We invite you to learn more about our work and welcome your insights as we continue to ensure AI is leveraged equitably and responsibly.

— — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — —

You can support our work by:

• Donating: No amount is too small. Every little bit can help us ensure underserved learners and educators are receiving the educational opportunities they deserve.
• Helping spread the word by engaging with us on LinkedIn, Instagram, YouTube, Tik Tok, and Facebook.
• Staying in the loop with our work by subscribing for monthly updates.

Is AI Fixing Education or Breaking What Matters Most? was originally published in Learning Equality on Medium, where people are continuing the conversation by highlighting and responding to this story.

The Plan, the People, the Impact

1) Five Years, 25 Million: Game On

In 2025, we launched a five-year strategy to reach 25 million learners by 2030. We’re primarily doing that by doubling down on our approach of: expanding access to offline digital learning, deepening the quality and impact of teaching and learning, and strengthening local capacity so organizations can own, lead, and scale what works for them. Grounded in community needs and with 12+ years of close collaborations with implementing organizations worldwide, this plan puts student-centered learning and offline edtech in service of teachers, learners, and local communities: from access to agency.

2) Welcoming new teammates

This year we were thrilled to add two incredible people to our team to help advance our mission:

• Susan Eyah — Learning & Impact Manager (Nigeria). Bringing deep experience leading MEL systems across humanitarian, education, and development programs (Mercy Corps, CARE International), Susan centers local ownership and evidence-based learning to elevate stories that move programs forward. She holds a Master’s in Information Management, and off the clock, enjoys curating epic Nigerian meals for family reunions.
• Kenneth Oroma — Program Specialist, Student-Centered Learning (Uganda). A former Teach For Uganda Fellow and Distinguished Fellow Award recipient, Kenneth has taught across Primary 1–7 and holds a Master’s in International Education & Development. He’s all in on unlocking every child’s potential,Man United-induced stress and all.

3) Open source helping open doors: Akshar Foundation (India)

Akshar Foundation is reimagining education — and tackling child labor — through a “learning and earning” model that blends student-centered academics with hands-on projects, vocational skills, and community service. Their implementation of Kolibri notably also shows how our open-source tool can be adapted and built upon for different organizational needs. In Akshar’s case, they use Kolibri to engage students in digital literacy, help teachers deliver truly personalized learning, simplify progress tracking for administrators, and even power a digital banking tool that supports safe earned income. It’s a powerful example of how adaptable tech can fuel inclusive, holistic education, and help children become agents of change. Watch the video and learn more.

4) DIY in action: Bondémi (Guinea-Bissau)

We love to learn about organic adoptions of Kolibri, which are made possible through our DIY Implementation model, supported by our comprehensive Edtech Toolkit. This year we had the pleasure of learning about Bondémi, an organization serving learners across the social, public, and development sectors in Guinea-Bissau. Led by digital capacity specialist Bruna Lobo, and supported by AJEP-GB ONG, UNDP, UN Human Rights, and UNICEF (with UN Peacebuilding funding), Bondémi uses Kolibri to democratize access to training content designed for local needs. Kudos to Bruna and the team championing equitable, inclusive digital learning!

5) Reshaping teaching and learning in rural Zambia: Impact Network together with PEAS (Zambia)

Academics Edward Banda and Jacob Khodowe, working with Impact Network together with PEAS, shared how Kolibri is reshaping teaching and learning across rural Zambia — boosting student engagement with interactive content and making lesson planning easier for teachers. With Kolibri now in 40+ government schools in Zambia, we’re proud to see educators and administrators adopting our tools and turning disconnected classrooms into places where learning thrives.

6) Equipping the next generation with the skills they need to thrive: Shoulder to Shoulder (Honduras)

This year, we watched our friends and long-time collaborators, Shoulder to Shoulder, continue to thrive in Honduras, reaching 20,000+ students across 100+ schools. Over the years, they have helped learners from grades 1–12 build digital literacy, reading, and STEM skills ( — even robotics!) — by adapting Kolibri to best fit their needs. Their video captures the impact beautifully.

7) Scaling entrepreneurship to fight extreme poverty: Village Enterprise (Uganda)

In 2025, Village Enterprise’s SPRINT program showed how smart tooling can multiply impact, and we were honored that Kolibri played a role. By pairing Kolibri-based training with tools like WhatsApp and DreamSave, SPRINT’s product scaling officers supported 5–10× more entrepreneurs, starting and graduating up to 600 businesses a year per officer (vs. 60–70 with traditional mentoring). Inspiring innovation in the fight against extreme poverty.

8) Hands Speak, Minds Grow: Family Educational Services Foundation (Pakistan)

In Pakistan — where education is often under-resourced and marginalization compounds for children with disabilities, — Deaf Reach, a program of the Family Educational Services Foundation, (FESF), is using Kolibri to support accessible, visual, and language-rich learning for learners aged 5–16, many with little or no prior schooling. At the Deaf Reach School (DRS) Lahore campus, the team, who are also the — official creators of Pakistani Sign Language, (PSL) — integrates Kolibri for foundational support across an art room, computer lab, and adult skills programs. Deaf Reach aims to expand to 50 DRSsS soon, with 50–70 more planned. They’re showing how offline-first technology can bring — bringing accessible, community-led learning to scale.

9) Radios, Routers and Reach: Elimu (Kenya)

We loved learning about a community-led Kolibri training by our friends at Elimu in Kenya. As part of their Offline eLearning Solutions (OES) program, Elimu paired Kolibri with long-range routers and Wi-Fi radios to reach remote schools. Over two days, educators dove into learner basics and admin tools: planning lessons, creating quizzes, grouping students, and using our curricular tool, Kolibri Studio, to build custom channels. Next up, selected schools received Kolibri-loaded servers with locally curated government content, thanks to Elimu Development Projects — bringing powerful offline learning even closer to the classroom.

10) Local Roots, Wider Reach: Blue Butterfly (Haiti)

Our partner Blue Butterfly works to ensure that Haitian children receive a holistic education that equips them to thrive in the contemporary world. The uniqueness of their approach lies in using evidence-based innovations like Kolibri and adapting them to the specific needs of the remote communities they serve. This year, they expanded Kolibri’s reach to even more local organizations, scaling the work to include grades 6 to 9, so more teachers can benefit from locally created resources and Kolibri’s UI in Creole.

AI that supports — not replaces — teachers

11) Offline, human-in-the-loop, equitable: our approach to AI

In 2025, everyone talked about AI in education, but hardly anyone focused on those who are most disconnected. At Learning Equality, we did. And we asked ourselves: how could AI actually help teachers and children without the internet it typically runs on? With 2.6 billion people still offline, we doubled down on a needs-driven, offline-ready, human-in-the-loop approach, co-designing with educators to ensure that AI supports (not replaces) teachers.

Practically, that means ensuring teachers have access to curriculum aligned lesson plans, digital resources, and assessments for quick pre/post tests. We also integrated AI literacy into Kolibri, so learners and educators could build AI skills safely and confidently. All based on the needs that we heard directly from our community, all intended to give teachers their time back. Check out for yourself!

12) From Months to Days: Curriculum-Aligned Content, Faster

In 2025 we continued to optimize our holistic semi-automated process that turns curricular documents into machine-readable taxonomies and recommends matching resources from a large OER library. New for this year: we advanced the process of AI digitization by working with The Curriculum Foundation to infer learning objectives from each section of the official South Sudanese STEM textbooks. In parallel, we’ve been applying our AI pipelines across the full Kolibri Library of 200K+ resources — transcribing captions, extracting structured metadata, and inferring learning objectives — to enable more detailed and precise curricular matching than any of our previous approaches. We’ve also been using this as an opportunity to generate synthetic datasets to feed into fine tuning the smaller LLMs that we’re working to integrate into Kolibri to support offline use cases.The result: teachers, students, and out-of-school youth in disconnected communities can access curriculum-aligned content faster.

13) Less Time Preparing, More Time Teaching

At UNESCO’s Digital Learning Week in Paris, our Co-Executive Director, Jamie Alexandre, joined Vodafone Foundation to share how AI-generated, teacher-co-designed lesson plans integrated with Kolibri supported refugee teachers in Egypt and South Sudan. The approach keeps resources curriculum-aligned and contextually relevant, reduces workload, and strengthens pedagogy, putting AI to work in service of teachers and expanding quality learning in crisis-affected communities. Learn more.

14) Equitable AI in Action: UNGA, New York

During this year’s United Nations General Assembly, we co-hosted a think-tank session called “Equitable AI in Action: A Collaborative Exchange for Low-Resource Contexts” with Associação Nova Escola, Rising Academies, and Luminos Fund at Education House (with support from Teach For All, HundrED, and Salzburg Global), sponsored by the Endless Foundation. At the event, nonprofit leaders, funders, practitioners, and educators gathered for lightning talks and roundtables to explore how AI can support teachers offline in low-resource contexts, while centering teacher agency. Three themes rang clear: keep equity and teacher agency at the center, share components and lessons to avoid duplication, and work together where approaches overlap. We’re deeply grateful to all who joined (including Hon. Conrad Sackey, Minister of Basic and Senior Secondary Education of Sierra Leone), and look forward to bringing insights into action in the new year.

From Challenge to Change with Flying Colors: Standing with Refugee Learners and Educators

15) Flying Colors: Building Foundations, Shared Practice, and Community

Our tech-enabled, project-based learning program for foundational and SEL skills building, Flying Colors, kept its momentum in Uganda with support from the Swarovski Foundation and our local partner, the refugee-led organization YIDA. Highlights from the year include a locally led, in-person Community of Practice in Palabek Refugee Settlement (with school visits and co-created classroom observation schedules) and a follow-up site visit, where members of our team had the chance to join YIDA in observing classrooms across refugee and host communities and meeting with teachers, school administrators, and government stakeholders to discuss scale, sustainability, and system strengthening. Shared meals (yes, the unforgettable goat roast) and a visit to a national park made the week as human as it was practical. All grounded in trust, shared practice, and community.

16) PBL Curriculum & Toolkit: open, practical, teacher-ready

Building on the needs our Flying Colors implementers shared with us, we created an openly licensed Project-Based Learning (PBL) Curriculum & Toolkit in the Kolibri Library, aligned to the P3 Abridged Ugandan curriculum. It puts 21 weeks of lesson plans, embedded PBL structures (driving questions, rubrics, presentations), parent/caregiver sessions, adaptable M&E tools, and locally relevant open resources (e.g., Ubongo Kids, African Storybook, Khan Academy) in one place. It also integrates Amal Alliance’s Colors of Kindness SEL activities and a full Training of Trainers pathway with CoP guidance and slide decks. Designed through participatory research with teachers and regional experts, it tackles real classroom constraints (mixed levels, limited time, scarce materials) so any educator can run PBL for foundational literacy and numeracy. No guesswork, no paywall.

17) From displacement to impact: James Gihoma’s Story

Refugee education has been core to Learning Equality’s work since its inception. On this year’s World Refugee Day, we honored the 122.1 million people displaced worldwide — 49 million of them children — by spotlighting James Gihoma, our partner and founder of the refugee-led organization YIDA Youth Initiative. By collaborating with UNHCR and Vodafone Foundation over the years, we’ve been able to advance our work developing human-in-the-loop, offline AI-powered tools, so that organizations like YIDA can ensure that even in crisis, children can learn, grow, and thrive. James’s story is one of resilience and community power, from displacement to impact.

Advocacy & Recognition: Momentum On The Global Stage

18) Building awareness, community, and advocacy across the globe

From Nairobi to Paris to Abu Dhabi, our team took the stage to share what equitable, student-centered learning can look like: rooted in a community-driven approach, powered by open, offline digital tools, and built with and for those we serve. Highlights included CIES 2025 (Chicago), World Schools Summit (Abu Dhabi), AI for Education Summit (Nairobi), mEducation Alliance Symposium (Washington, DC), EDTECHWEEK (New York), FiRe and ASU+GSV Summit (San Diego), the Google Summer of Code Mentor Summit (Germany), the Social Innovation Summit (San Francisco), and the International Fundraising Congress (Netherlands). Each one a chance to learn, connect, and move the field forward with equity at the center.

19) Momentum, Validated

In 2025, our work was recognized across the sector. We won the Dataset Prize in the 2025 Tools Competition (Renaissance Philanthropy) for a tool that uses anonymized Kolibri search queries to improve discovery while creating much-needed datasets from underrepresented communities for policy and education researchers. We were selected as one of 11 Innovators to Watch in Jobs for the Future’s Market Scan for transformative edtech in prison learning programs, recognizing Kolibri’s role in expanding opportunity for learners in U.S. correctional facilities, and we were also named a Top 10 finalist for the inaugural Global EdTech Prize (T4 Education, with Owl Ventures and Digital Promise). Lastly, we were honored to join the Global Schools Forum, a collaborative community of non-state organizations working to improve education at scale for children in low- and middle-income countries, and have Flying Colors be included in the HundrED 2025 Global Collection, with specific features as part of their “Child-Centered Learning” and “Wellbeing in Schools” spotlights. We are honored to share each of these with inspiring peers advancing technology in service of learners.

Fly, Kolibri, Fly…

20) Soaring Higher

Number of newly known Kolibri installations: Approx. 54,000 (the year hasn’t ended yet!)

Note that this number reflects only known Kolibri installations (which can serve one individual learner or support an entire school) that have gone online, whereas the majority of Kolibris never go online and are often downloaded once and installed onto many offline devices, which we can’t track. Still, it’s a joyful signal: Kolibri continues soaring to greater heights, opening the world of learning for even more teachers and students, everywhere.

21) Streamlined for teachers

This year, our product team gave teachers working with Kolibri a much more streamlined experience. Based on user feedback, the v0.18 family redesigned coach navigation with simplified tabs: “Lessons,” “Quizzes,” and “Groups,” making it quicker and more intuitive to find and manage key teaching tools. Quiz creation also changed, giving educators greater customization and control by allowing them to handpick questions and choose when learners see their results. If that wasn’t enough, the release also improved content discovery with metadata filters for resource type, language, and learning level, and enhanced tagging, so it’s easier to find the right materials fast. In short: less time setting up, more time teaching!

22) Shaping Kolibri Studio together

Once more in 2025, our community came together to help shape a new generation of features in Kolibri Studio. Through quick feedback calls with our design team, creators and implementers showed us how they customize assessments, track learner progress, navigate the public library, and where there was friction. So we turned that insight into action with AI-powered search recommendations in Kolibri Studio. Paired with thorough metadata audits and updates, discovery of relevant content is getting sharper and Studio workflows becoming noticeably more efficient and intuitive. We’re excited to share it with you in the new year.

Rolling the Credits: Heartfelt gratitude to our supporters

23) Built with Community, Powered by Volunteers

In 2025, our volunteer developer community donated time and talent to strengthen Kolibri and Kolibri Studio for teachers and learners around the world. On behalf of Learning Equality and everyone who relies on our tools, thank you for helping Kolibri fly farther!

With deep gratitude to:
Aadarsh Mahesh K, Abhishek Punhani, Abhishek Yadav, Adib Mubarak, Anas K., Anoushka Jha, Arunima Goel, Arunit Chakraborty, Chetan More, Dorna Raj Gyawali, Eddie Liu, Eshaan Aggarwal, Fidal Mathew, Garvit Singhal, Gautam Manchandani, Jason Altekruse, Jayesh Savaliya, Joseph Avila Alvarez, Karan Kumar Das, Kartikay Singh Pundir, Kshitij Thareja, Lokesh Varma, Mukul Rana, Nandini Saagar, Nihal Shinde, Prashant Thakur, Rafal Paradowski, Rishabh Jain, Ronak Raj, Sahil Sinha, Sanchit Maini, Shivam Daksh, Shobh Raj, Shruti Yadav, Sujai Kumar Gupta, Sukhvir Singh, Tushar Verma, Viraj Jaiswal, Vivek Pathak, Yash Kumar Singh, Yeshwanth Munagapati, and those we only know by their GitHub handles: @Divyanshi750, @kart-u, @malviya-rajveer, @vaibhav-if.

24) Google Summer of Code: 5x5 in 2025

For the 5th year in a row, we had the honor of being selected as a Google Summer of Code mentoring organization, with 5 projects, tackled by 5 brilliant individuals:

Habiba Ayman, Winny Chang, Shruti Yadav, Jakub Komárek, and Dimitrios Mylonas.

We thank each of them for their time and talent helping us enhance our tools so that learners and educators everywhere, regardless of connectivity, can enjoy a quality digital learning experience.

25) That Simple Equation: Intention + Commitment = Impact

We extend our heartfelt thanks to so many of our supporters including Woven Foundation, Patrick J. McGovern Foundation, Swarovski Foundation, Equinix Foundation, Vodafone Americas Foundation, Vodafone Foundation, Endless Foundation, Hewlett Foundation, Hewlett Packard Enterprise , Salesforce, Kung Guerra Foundation, and Fast Forward for their support in 2025. Collaborations like these remind us that everyone has a role to play, and when organizations show up with intention, their investment turns into impact.

— — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — —

Reading this back, the throughline is clear: when the tools are open, the design is grounded in needs, and the leadership is local, learning can travel — even when connectivity doesn’t.

This year was proof that progress isn’t a solo; it’s a chorus, built in classrooms, codebases, community gatherings, and long conversations about what equity in learning really looks like.

Thank you for being part of it. As we head into 2026, we’re carrying the same commitment forward: expanding access, deepening quality, and strengthening local capacity.

If you can, share this recap, spread the word about our work, or consider a gift to help us keep opening the world of learning for everyone, everywhere. 💛

Our Year in Review: 2025 was originally published in Learning Equality on Medium, where people are continuing the conversation by highlighting and responding to this story.

Kolibri v0.19 makes user management easier for program administrators, from navigating the dashboard to creating accounts and managing users in bulk. By saving time on administrative work, administrators can focus on what matters most: supporting their teachers and learners.

What’s changing:

Based on feedback from our community and implementing partners, we’ve significantly improved how users can be managed. Our latest changes reduce the number of repetitive tasks required for class enrollment and unenrollment of learners, and general account management. This makes both setting up new programs and year-over-year user management simpler and more efficient.

Previously, many updates had to be made on a per-user basis, one account at a time. Administrators can now perform actions on multiple users at once, reducing the time needed to manage users at scale. Key workflows include:

• Bulk user selection and actions: Allows admins to select multiple users and apply actions to all of them at once, including enrollment, removal from classes, and deletion
• Learner-only device management: Allows admins to remove learner-only devices users from a device and import them outside of the setup wizard workflow
• Improved user creation workflow: Admins can now enroll a user in one or more classes during user account creation

Here’s a sneak peek at Kolibri v0.19:

But wait, there’s more!

Need to enroll all learners the same year in separate classes for Math, Science, Reading, and History? Now, admins can also create copies of an existing class, to easily duplicate grouped learners and coaches in the same cohort. Make one class, copy and update the class name, and your setup is complete!

What’s Next

Looking ahead, our roadmap this year includes the much-anticipated launch of the Community Library, which will boast content locally created by our users, as well as content channels aligned to local curricula. We will also update the import and discoverability experience of channels (including Community Library channels!) in Kolibri, to help make it easier for administrators to set up Kolibri devices quickly and efficiently.

We will also be launching resource upload in lessons, allowing Kolibri Learning Platform users, for the first time, to import their own resources directly into the platform, without having to go through Kolibri Studio.

These updates add up to less setup and more learning: local channels at your fingertips, content that’s easier to find, and your own resources right inside lessons.

We’re excited to get to work… and even more excited to see what you do with it! As always, we invite you to join us in making education accessible to all. You can support our work by:

• Donating: No amount is too small. Every little bit can help us ensure underserved learners and educators are receiving the educational opportunities they deserve.
• Sharing your Kolibri story with us via the Community Forum.
• Helping spread the word by engaging with #KolibriFly on LinkedIn, Instagram, Bluesky, YouTube, Tik Tok, and Facebook.
• Staying in the loop with our work by subscribing for monthly updates.

New Year, New Kolibri: v0.19! was originally published in Learning Equality on Medium, where people are continuing the conversation by highlighting and responding to this story.

Today there are 12.4 million school-aged refugee children, of which 46% are missing out on an education. That’s over 5.7 million children not in school (source: UNHCR Education Report). Those who do attend encounter teachers who often face overcrowded classrooms, multiple languages, scarce resources, and extremely limited preparation time.

It is in this context that Vodafone Foundation and Learning Equality are leveraging AI to strengthen pedagogical support for refugee teachers, complement teacher training and provide ready-to-use lesson plans with an aim of better equipping teachers in classrooms with limited resources.

We are building this into the Instant Network Schools (INS) program — a partnership between Vodafone Foundation and UNHCR, the UN Refugee Agency, that brings quality digital education to refugee-hosting schools. INS currently operates across six African countries. To date, we have reached more than 382,000 students and over 6,800 teachers, so there is potential for AI to have a huge impact on the quality of education in these communities.

We started our AI integration pilot in Egypt and South Sudan. This initiative builds on years of foundational work by Learning Equality, Vodafone Foundation and UNHCR to improve the process of bringing together relevant, supplemental content that can support teaching and learning.

Since its inception, Learning Equality has focused on making high-quality educational resources accessible offline and developing tools to align them with national curricula. In low-resource contexts, this alignment is essential: it ensures that teachers and learners can discover and engage with content that is both relevant and accessible.

Guided lesson plans are the natural next step — providing structured guidance and classroom activities alongside organizing these aligned resources into structured, pedagogically sound sequences that teachers can use directly in the classroom.

Four lessons from our pilots in Egypt and South Sudan:

1. Teacher capacity is the most powerful lever for student learning.

Research shows that teacher knowledge can explain up to 37% of the variation in student achievement. Strengthening refugee teachers’ ability to engage learners in overcrowded classrooms is one of the most effective investments we can make.

2. Guided lesson plans are an impactful entry point.

Aligned to national curricula and integrated with life skills, these plans are built on active learning methods and come with ready-made resources, such as presentations, quizzes, and interactive activities. This saves teachers valuable preparation time while boosting their confidence and classroom engagement.

Just as importantly, the lesson plans provide a meaningful way for teachers to adopt technology-enabled, student-centered approaches. For many refugee teachers, who often have little access to training or ongoing professional development, this structured guidance acts as a bridge, helping them confidently deploy active pedagogy in their classrooms.

3. Human-led inputs are critical to drive AI.

In Egypt, over 780 lesson plans were researched, co-designed, and piloted with teachers, then refined into a standardized structure linked to the INS Teaching for Life framework.

In South Sudan, we followed another approach: documenting the human “thought process” behind lesson planning by capturing the sequencing of decisions, instructional approaches, and references used.

Both of these pilots are being evaluated by humans, and can inform future improvements in lesson plan design. This not only makes today’s lesson plans usable, but also creates structured inputs to inform how AI can replicate high-quality pedagogical decision-making at scale.

4. AI is an enabler, not a replacement.

As of now, AI supports transcription, translation, and structuring of content. Looking forward, it can accelerate lesson plan production across geographies. But human expertise in pedagogy and cultural context remains essential to ensure quality and relevance.

Impact in the classroom

Teachers themselves are already reaping the benefits of having contextually-specific lesson plans. As one INS teacher reflected in Egypt:

“With the new guided lesson plans and digital resources, I can spend less time preparing and more time engaging my students. Even in a crowded classroom, I see more participation and excitement for learning.”

By embedding active, student-centered methods into every plan, this approach is also helping teachers deploy new technology-enabled strategies that might otherwise remain out of reach. Teachers not only save time but also gain a practical pathway to more engaging and participatory classrooms.

Looking ahead

As AI evolves, these pilots demonstrate the potential of combining human expertise with AI efficiency to deliver lesson plans that are both scalable and context-appropriate. For teachers, this means more confidence and less preparation time. For students, it means more engaging lessons that build knowledge as well as critical life skills.

Vodafone Foundation and Learning Equality are committed to continuing to explore how AI can drive the creation of educator-support materials, including for use across more INS countries. Guided by a pedagogy-first approach and continuous teacher feedback, our goal is not to replace teachers but to empower them, and to bring out the best in every learner, even in the most challenging circumstances.

Less time preparing, more time teaching: How AI is helping refugee teachers was originally published in Learning Equality on Medium, where people are continuing the conversation by highlighting and responding to this story.

At Learning Equality, we serve those without access to quality education or reliable connectivity because traditional edtech often fails in these environments.

Today, our approach, products, and tools address multiple gaps:

• A global learning crisis where 70% of 10-year-olds cannot read and understand a simple text.
• An implementation gap, where teachers are expected to lead digital instruction but aren’t given tools that reflect their needs or constraints.
• A relevance gap, where educational content isn’t aligned to local curricula or delivered in local languages.
• A systemic equity gap, where disconnected communities are excluded from innovations in teaching and learning.

We develop open educational tools that foster agency for learners, teachers and organizations.

Thanks to our solutions, over 13 million learners in underserved communities across 220+ countries and territories now have access to transformative digital tools. No connectivity required.

We also recognize that technology alone isn’t a silver bullet, so we approach edtech differently: We design it for the contextual realities for when the Internet is limited or does not exist, leveraging it to foster student-centered learning practices.

Why student-centered learning?

Because in many underserved contexts, traditional approaches still hinge on teacher-led instruction with minimal student participation. A student-centered model empowers learners, giving them choice, voice, and deeper engagement in shaping their educational journey. This philosophy, paired with intentional offline-first edtech design, ensures that even in disconnected environments, every learner can be an active agent in their own growth, leading to better learning outcomes.

Beyond that, our focus is honed in on providing organizations — through direct assistance or our DIY model — with what they need to ensure that the future of learning doesn’t belong only to those already connected.

Over the past twelve years, we’ve seen what’s possible for learning when dedicated communities, educators, and organizations come together around a shared vision. From classrooms in rural villages to learning centers in refugee camps, we’ve witnessed offline-first technology unlock opportunities where few believed it could. This experience has laid a strong foundation, and now, we’re ready to build on it with even greater ambition and clarity.

That’s why we’ve set ourselves an ambitious goal: Reach 25 million learners by 2030.

We’re going all in on realizing our vision of a world where every learner can thrive. Because in five years, learning will look completely different shaped by AI, and those who have power over information. And if we’re not intentional, the same students being left behind in today’s classrooms will be left out of tomorrow’s world.

“Looking ahead, our strategy for the next five years is both a continuation and an evolution. We know that access alone isn’t enough and we’ve always focused on enabling student centered learning. And now we see an opportunity to deepen the quality of learning experiences and focus more on student centered learning and working more closely with organizations to drive local change but also accelerate learning outcomes at scale”.

— Navya Akkinepally, Co-Executive Director

Our 5-Year Strategy

Our new strategic plan takes what we’ve learned over the years and doubles down on three fronts: expanding our reach, deepening our impact and the quality of learning and teaching we enable, and strengthening local capacity for the organizations and communities we serve. Here’s how we’re getting there:

We’ll continue doing what we do best: innovating with Kolibri — our open product ecosystem — to promote offline-first edtech and student-centered learning. As education evolves with new technologies, we’ll keep centering those furthest from opportunity, ensuring underserved learners and educators benefit directly from AI. We’ll focus on building digital skills and AI literacy in disconnected communities and invest in R&D in emerging technologies. Our work will remain grounded in feedback from our user community, while we deepen collaboration with implementing partners to strengthen Kolibri, enhance organizational support, and drive meaningful learning outcomes. As part of this, we’re investing in rigorous education research by co-creating learning agendas that generate actionable evidence on what drives teacher practice change and student progress in low-resource settings, while contributing to the broader education research and policy landscape.

We will also continue to focus on building foundational and socio-emotional skills for young learners through tech-enabled project-based learning. This approach addresses the issue of learning poverty by helping children develop basic literacy and numeracy skills through play and problem-solving activities. We’ve seen, first-hand, that this method works, and we remain committed to deepening its impact on learners and educators globally.

We are going to accelerate our progress toward our product vision of aligning content to every country’s curriculum by supporting local content creation, developing tools to digitize and align standards, and integrating content recommendations into our offline product. This work leverages advances in GenAI and our machine learning algorithms to drive edtech adoption, making it easier for underserved communities to adopt tools that fit their unique educational goals.

Lastly, we’re going to launch something new: a collective of organizations united by a shared mission of advancing student-centered learning through offline edtech. The Collective will promote collaboration, peer learning, and local ownership — essential for building strong partnerships and scaling sustainable adoption. By working together, members of the Collective will strengthen their ability to support educators, expand local content creation, and improve learning outcomes. They’ll also gain deeper insights into unique organizational needs and impact as they build a stronger evidence base to support broader adoption of offline-first edtech and student-centered learning practices.

This new initiative is also aligned with our recent shift to a distributed leadership, and our ongoing commitment towards a more community-proximate organizational model.

With the word, our Co-Executive Directors

On a candid chat, our Co-Executive Directors, Navya Akkinepally, Lauren Lichtman, and Jamie Alexandre share more about our 5-year Strategy: Why it matters, what it means for learners and partners around the world, and how we’re turning vision into action.

Watch the video below to hear their insights firsthand.

How You Can Get Involved

While this strategy lays the path forward — the real impact only happens when it’s put into practice. Everything we’ve achieved thus far has been shaped by the people who use our tools, adapt them, and bring them to life in their communities. That’s why we’re inviting you to join us in turning vision into reality.

Our journey forward depends on those who believe, like we do, that every learner deserves a seat at the learning table — no matter where they live or what resources they have.

We’re more excited than ever about what the future holds, and we invite you to join us in this journey:

• Donate! No amount is too small. Every little bit can help us ensure underserved learners and educators are receiving the educational opportunities they deserve.
• Let us know about other initiatives that we should be aware of as we embark on this new Collective by sending us an email at info@learningequality.org.
• And if you’re already using Kolibri, share your story with us via the Community Forum.

You can also watch #KolibriFly on LinkedIn, Instagram, Bluesky, YouTube, Tik Tok, X, and Facebook, and stay in the loop with our work by subscribing for monthly updates here.

From Access to Agency: Embarking on a 5-Year Journey to Reach 25 Million Learners was originally published in Learning Equality on Medium, where people are continuing the conversation by highlighting and responding to this story.