LearnID | Devpost

Inspiration

Students across campuses struggle to remain engaged in lectures due to not conforming to the cookie-cutter studying routines of today's universities. We realized that "learning styles" aren't just preference. They are a core part of a student’s Cognitive Identity. When a student says "I'm bad at math," it's usually because the math isn't speaking their brain's native language. We built LearnID to turn the "one-size-fits-all" lecture into a conversation that adapts to the student.

What it does

LearnID uses AI to build individualized learner profiles for users, identifying how a student most efficiently retains educational concepts to generate custom lecture slides designed to maximize user engagement and focus. Students evolve, and so does LearnID, constantly supervising users' focus levels throughout a lecture to tweak and improve generative approaches.

How we built it

Frontend: Built with Next.js and Tailwind CSS for a responsive, accessible UI. We used Recharts to visualize student identities and FaceAPI.js for client-side focus tracking.
Intelligence Layer: We integrated the Gemini API as our core reasoning engine. It analyzes raw behavioral data (dwell time, scroll speed, focus drops) to determine the optimal content format for the next chapter.
Generative Engine: For visual learners, we utilized the Manim (Python library) on the backend to programmatically render high-quality mathematical animations. For text learners, Gemini dynamically restructures complex theories into bulleted summaries and worked examples.
Backend & Data: A FastAPI server handles the logic, while MongoDB Atlas serves as our flexible data backbone, storing event streams, user profiles, and adaptation logs in a scalable cloud environment.

Challenges we ran into

Real-time Adaptation: It was difficult to synchronize the computer vision focus-tracking with the slide-rendering engine without creating lag. We optimized this by using background workers to process behavioral data asynchronously, ensuring the learning experience remained fluid.
Cold-Start Problem: Defining a student's identity from scratch is a significant hurdle. We addressed this by implementing a "baseline lecture", a fixed-format introductory session designed to gather enough high-quality initial data for Gemini to make its first high-confidence adaptation.
LLM Orchestration & Quotas: As this was our first time working deeply with LLM APIs, we faced a steep learning curve in prompt engineering to ensure consistent, structured JSON outputs for our frontend. Additionally, we had to navigate strict API rate limits and token quotas. We overcame this by implementing a caching layer for common content transformations and optimizing our prompts to be as token-efficient as possible without losing pedagogical nuance.

Accomplishments that we're proud of

Dynamic Modality Switching: Successfully harnessing the Gemini API to dynamically develop static textbook-like explanations for text-first learners, and Manim renders for animated graphics for visual learners.
Behavioral Synthesis: Building a system that doesn't just ask what you like, but proves what you need by correlating engagement time with quiz performance.
Identity-Centric UI: Creating a dashboard that empowers students to understand their own neurodivergence as a strength rather than a hurdle.

What we learned

The Power of Unstructured Data: Using MongoDB Atlas taught us the value of document-based schemas when dealing with unpredictable behavioral events like eye-tracking coordinates and clickstreams.

What's next for LearnID

Actionable Dashboard Insights: We plan to evolve our student dashboard from simple tracking to "Metacognitive Coaching," providing students with specific strategies—like optimal study times and focus-restoration techniques—based on their unique data.
Seamless LMS Integration: Our goal is to develop a Canvas and Moodle plugin, allowing professors to "Identity-proof" their existing curricula with a single click. This turns static document repositories into living, adaptive environments.
Empathetic Reasoning: To deepen our AI’s effectiveness, we aim to allow students to securely opt-in to sharing broader context such as language background or specific neurodivergent diagnoses. This enables the Gemini-powered engine to be more empathetic, adjusting tone and complexity to meet the student's lived experience.
Collaborative Learning Cohorts: We want to bridge the gap between individual identity and community. By connecting students with similar "Learning Fingerprints," LearnID can facilitate peer-study groups where everyone is already speaking the same "brain language."