AI Tutor

A personalized AI tutor that learns user’s progress and autonomously suggests material and schedules.

Comment

Smart Agent for Personalised Learning & Mentorship

Inspiration

As engineering students passionate about both AI and education, We often felt overwhelmed by the vast amount of learning content available online. There was no personalised system that could:

  • Track our progress,
  • Recommend precise next steps,
  • Adjust dynamically to our availability and skill level,
  • Or give us an intelligent mentor-like experience.

We envisioned a personal AI mentor, not just a chatbot, but a full-stack learning companion that evolves with us. With the rise of LLMs, vector databases, and personalised recommendation systems, this vision finally felt achievable. The Bolt Hackathon was the perfect environment to bring it to life.

What We Learned

This project taught us how to bridge multiple advanced AI tools into one seamless, end-to-end platform:

  • We explored Retrieval Augmented Generation (RAG) with Pinecone and Mistral to enhance AI's contextual memory.
  • We learned to use TensorFlow Recommenders to dynamically suggest next learning steps based on behaviour.
  • We built a smart scheduler that auto-plans a user’s week based on availability, difficulty, and prior history, simulating reinforcement learning-style feedback.
  • We gained deep experience with Bolt.new, creatively using its components to simulate a robust, real-time, AI-enhanced full-stack app.

How We Built It

Tech Stack:

  • Frontend: Bolt.new (React abstraction)
  • Backend/API: Bolt logic blocks + Firebase Functions
  • Auth & Database: Firebase Auth + Firestore
  • LLM Integration: Mistral API
  • Memory Search: Pinecone for RAG + long-term interaction embeddings
  • Recommendations: TensorFlow Recommenders (trained externally, used in inference logic)
  • Deployment: Netlify (for bonus), Entri custom domain

Core Features:

Feature Description
AI Mentor Chat Mistral-based chatbot with contextual answers based on the user's progress and goal
RAG Content Fetcher Embedded GitHub/StackOverflow content indexed in Pinecone, fetched for relevant topics
Autonomous Schedule Planner AI plans weekly tasks based on user availability and preferences
Smart Recommendation Engine Suggests next topics using TensorFlow Recommenders
Dashboard & Tracker Visualises progress, completion, and learning trajectory

Key Implementation Steps

  1. Created UI Skeleton in Bolt.new : Navigation layout with Dashboard, Chat, Schedule Planner, and Settings.
  2. Integrated Firebase Auth : Seamless login/signup and real-time progress sync via Firestore.
  3. Built Mentor Chat with Mistral : Connected OpenAI API with dynamic prompts using user’s current progress and goals.
  4. Embedded RAG with Pinecone : Vectorised learning resources, enabling context-aware resource retrieval.
  5. Added Smart Scheduler : AI agent planned weekly learning blocks considering the user's time availability and learning path.
  6. Trained TensorFlow Recommender : Fed learning data to suggest the next best topics for mastery.
  7. Deployed on Netlify : Custom domain configured with Entri for public demo and bonus eligibility.

Challenges We Faced

Challenge How We Tackled It
Real-time AI Feedback without Lag Optimized prompt structure + used Pinecone's similarity search for pre-context
RAG hallucinations from Mistral Introduced citation fallback from indexed content
Personalization vs. Overfitting Used random exploration factor in recommendations
Dynamic scheduling logic Created rule-based scheduler with task weights based on topic complexity
Integrating multiple services within Bolt Segregated logic cleanly using Bolt functions + Firebase logic bridges

Why This Project Matters

  • EdTech + AI is the perfect union for impact: democratizing access to quality mentorship.
  • It leverages AI agents, scheduling intelligence, recommender systems, and modern UI to mimic real human mentorship.
  • It’s modular, scalable, and a great fit for real-world implementation beyond the hackathon.

Final Thoughts

This project was more than a hackathon build; it was an experiment in autonomous, personalised learning. We plan to extend this into a production-grade tool, collaborating with educators and students to scale intelligent mentorship.

Built With

Share this project:

Updates