MentoLab

Front Page

Inspiration

I’ve always noticed how differently people learn to code and understand AI. Some learn by experimenting, some by debugging, others through patterns or visuals. But most platforms force everyone into the same path. I wanted to build something that unites these differences and turns them into a shared advantage. That idea became MentoLab—a mentor-like system that learns from the diversity of how people learn and uses those insights to guide each new learner.

What it does

MentoLab is a personalized AI platform that teaches coding and machine learning through adaptive, community-informed challenges.

It: Uses DRL to choose the optimal next challenge for each learner. Uses a Bidirectional RNN to model a user’s evolving learning style, debugging habits, and pace. Builds a GNN-based similarity graph that lets the system transfer effective learning sequences from users with totally different backgrounds. Converts anonymized mistakes from others into debugging tasks for new learners. Explains the ML logic behind each recommendation, so users learn AI literacy while learning to code.

How we built it

I combined three ML components: A policy-gradient DRL agent for dynamic task sequencing A GRU/BiLSTM RNN to track learning signatures A GNN similarity graph to share helpful learning paths across diverse users I also built a RAG task generator that rewrites real user errors into practice problems, a clean UI with an “AI reasoning panel,” and a Python + Node backend with vector embeddings and fast RL inference.

Challenges we ran into

Stabilizing DRL rewards so difficulty scaled naturally Making AI decisions transparent without overwhelming users Preserving privacy while using community mistakes constructively Getting the GNN to reflect learning behavior, not superficial traits

Accomplishments that we're proud of

A fully integrated DRL + RNN + GNN learning engine Real-time adaptive paths that update after every user action A system that unites different kinds of learners through shared behavior patterns Turning community errors into high-quality learning opportunities

What we learned

Learning process matters more than learning content Diverse user trajectories create the strongest personalization signals DRL excels when rewards are tied to comprehension, not correctness Transparent AI dramatically boosts user trust

What's next for MentoLab

A Path Remix tool for visualizing and editing learning paths More advanced ML topics and interactive tasks Collaborative debugging powered by similarity graphs Educator tools for shaping DRL-driven lesson structures Privacy-preserving, scalable training using global learning data