🎓 LearnAI — Personalized AI Tutor
An adaptive learning platform that teaches AI/ML the way you need to learn it
It diagnoses what you know, finds where you struggle, adapts difficulty in real time, and rebuilds your learning path after every answer.
📌 Table of Contents
❗ The Problem
Traditional AI/ML education treats every learner identically. A developer with two years of Python and a complete beginner get:
- the same lecture videos and explanations,
- the same math depth (either too opaque or too dumbed down),
- the same practice problems and fixed linear curriculum,
- the same pace.
When a learner fails a fundamental concept like the Bias–Variance Tradeoff, most platforms just mark it wrong and move on. Gaps compound until the learner gets overwhelmed and drops out.
💡 The Solution
LearnAI turns learning into a closed feedback loop. It continuously estimates each learner's knowledge state per concept, detects weaknesses immediately, calibrates question difficulty, and restructures the roadmap, while an AI tutor explains every concept at the learner's exact level.
Same question, different learner, different answer.
Ask "What is gradient descent?" as a beginner and you get a foggy-hill analogy with 5 lines of commented Python.
Ask it as an advanced learner and you get the update rule, a Taylor-expansion derivation, and SGD/momentum/Adam trade-offs.
🌐 Live Demo
Demo personas (switch from the profile switcher in the app):
| Persona | Level | Goal | Profile |
|---|
| Alex | Beginner | Learn ML from scratch | Basic Python, weak on matrix shapes & variance |
| Akshat | Intermediate | Become an ML Engineer | Strong Python, weak on bias–variance & gradient descent |
| Elena | Advanced | Build GenAI apps | Senior DS, weak on vector-DB chunking & semantic drift |
⏳ The backend runs on Render's free tier, so the first request after idle can take 30–60 seconds while it wakes up.
✨ Key Features
🧠 Adaptive Intelligence (ML Backend)
- Learner Model: per-concept mastery, confidence and trend tracking across 26 AI/ML concepts.
- Weakness Detector: priority-scored weaknesses (HIGH / MEDIUM / LOW) with human-readable reasons and tutor alerts.
- Adaptive Difficulty Engine: picks Easy / Medium / Hard from the last 5 answers, time taken and confidence, not just the score.
- Recommendation Engine: builds a personalized learning path over a prerequisite DAG; skips what you know and injects revision for what you don't.
- Cold-Start Profiler: turns a 6-step onboarding questionnaire into a baseline knowledge vector and a first roadmap.
🤖 AI Tutor
- 9 learning modes: explanation, simplify, example, code, practice, hint, evaluate, revision, path.
- Level-calibrated depth: analogies for beginners, sklearn + loss functions for intermediates, paper-level mechanics for advanced learners.
- Hinglish support: replies in the learner's own language register.
- Verified checkpoint questions inside explanations.
- Production guardrails: prompt-injection defence, off-topic redirect, answer-leak prevention in hints, distress handling.
📝 Adaptive Practice
- LLM-generated questions targeted at detected gaps, with varied real-world scenarios.
- Independent answer verification: a second model solves every question; wrong or ambiguous ones never reach the learner.
- "Why this question?" explanation for every question, built from the learner's real data.
- Verified question pool so practice keeps working even under AI rate limits.
This is the beginning of the README. Read the rest at the original. Read the full README ↗
