OnLearn | Devpost

Your personal AI tutor landing page
Modules for DSA and Data Science(Will update further in future)
Your personalized learning plan. Updates dynamically as you progress.
This is your learning interface. The concept-tutor agent teaches pre-requisite concepts in a socratic approach.
Your integrated coding environment. The lab-mentor agent guides you throughout to write your code.
Your course structure to view the complete structure and navigate between the topics
Your progress tracker

Inspiration

While preparing for internships at IIT Roorkee, we found ourselves stuck in the same loop millions of students face—solving endless DSA problems, jumping between YouTube videos and one-size-fits-all articles, yet still feeling directionless. According to our research, 4M+ engineering students in India and 50M+ technical learners globally struggle with the same “tutorial hell.” We realized the real issue isn’t the lack of content but the lack of personalized, real-time guidance. That frustration, combined with our curiosity for agentic AI systems, inspired us to build the adaptive tutor we always wished we had.

What it does

OnLearn is an AI-powered adaptive tutor that identifies conceptual gaps, analyzes mistakes, and dynamically reroutes a learner’s path. It provides Socratic, conversational explanations and adjusts difficulty based on pace and retention—acting like a 1:1 intelligent mentor rather than a static content platform. It bridges theory and coding practice through an integrated IDE, ensuring learners move from confusion to clarity through personalized pathways rather than one-size-fits-all content.

How we built it

We built a multi-agent system where a Master Agent plans learning paths, a Concept Tutor teaches using structured lessons, and a Lab Mentor supports coding with sandboxed execution. Progress, context, and memory persist through a stateful backend backed by structured prerequisite mapping and concept dependencies.

Challenges we ran into

Designing meaningful personalization was difficult—early versions felt generic. Modeling prerequisite relationships, building persistent learning memory, and ensuring stable multi-agent coordination were difficult. Maintaining low inference cost while delivering real-time adaptive feedback required extensive iteration and pipeline optimization. Modeling student mastery, tuning weights, and crafting human-like pedagogy demanded multiple iterations. Ensuring smooth transitions between conceptual learning and coding also required significant system engineering effort.

Accomplishments that we're proud of

We built a fully functional adaptive tutor with conversational understanding, real-time code evaluation, deep personalization, and persistent learning memory. Over 150 students tested early versions and reported faster clarity, reduced confusion, and improved confidence—validating both demand and the effectiveness of our adaptive learning design.

What we learned

We learned that learning is nonlinear, and effective personalization requires combining diagnostics, memory, and adaptive sequencing. Real pedagogy isn't just correct answers—it’s guiding reasoning. We also gained strong insights into designing agent workflows, modeling mastery, optimizing LLM costs, and building user-centric educational experiences.

What's next for OnLearn

We’ll expand beyond DSA and Data Science, into broader STEM subjects. Next steps include campus partnerships, and deeper mastery modeling using probabilistic graphs. Our long-term vision is to make OnLearn a global AI-native learning ecosystem delivering truly personalized education at scale.