Personalized learning, powered by LLMs, RAG, and autonomous agents. A full-stack AI application demonstrating production-grade architecture across web and mobile.
UltraLearning is a three-tier adaptive learning platform that uses a full AI stack — Large Language Models, Retrieval-Augmented Generation, and autonomous agents — to deliver a personalized educational experience that adapts to each learner's pace, strengths, and goals.
Built to demonstrate how modern AI primitives (LLMs, RAG, agents, vector search, real-time streaming) can be composed into a coherent, production-architected application across backend, web, and mobile layers.
 |
 |
 |
 |
 |
 |
 |
 |
View all mobile screenshots (16)
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
View all web screenshots (16)
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
┌─────────────────────────────────────────────────────────┐
│ CLIENT LAYER │
│ Next.js Web App │ React Native Mobile │
│ (SSE Streaming) │ (iOS & Android) │
└────────────────────┬────────────────────────────────────┘
│ REST / WebSocket
┌────────────────────▼────────────────────────────────────┐
│ API LAYER │
│ Flask (Python 3.11) │
│ Auth (JWT) │ Agent Orchestration │ RAG Engine │
└──────┬──────────┬──────────────┬────────────────────────┘
│ │ │
┌──────▼───┐ ┌────▼─────┐ ┌─────▼──────────────────────┐
│PostgreSQL│ │ Milvus │ │ LLM Layer │
│(pgvector)│ │ Vector │ │ LangChain · MCP · Llama │
└──────────┘ └──────────┘ └─────────────────────────────┘
| Layer | Technology |
|---|---|
| Backend API | Flask, Python 3.11 |
| AI Orchestration | LangChain, Anthropic MCP |
| RAG Pipeline | Milvus vector DB, sentence-transformers (all-MiniLM-L6-v2) |
| LLM Inference | Llama (local), OpenAI |
| Web Frontend | Next.js 14, TypeScript, TailwindCSS, SSE streaming |
| Mobile | React Native (iOS & Android) |
| Auth | JWT |
| Database | PostgreSQL |
| Deployment | Docker Compose, Nginx |
🤖 AI-Powered Learning — Personalized content generation, summaries, and explanations via LLMs, adapting to each learner's history and performance.
🧠 RAG Document Intelligence — Upload your own PDFs and notes; they're chunked, embedded, and integrated into your personal learning context for grounded Q&A.
🤖 Autonomous AI Agents — Multi-agent system for research, real-time Q&A, and proactive learning assistance via MCP-connected context.
💬 Real-Time AI Tutoring — Token-streamed chat sessions via SSE with AI tutors and fellow learners.
🃏 Adaptive Flashcards — Spaced-repetition flashcards that adjust difficulty based on individual performance signals.
🏆 Gamification & Leaderboards — Points, badges, streaks, and competitive leaderboards to sustain engagement.
📊 Progress Dashboard — Topic-level mastery tracking with AI-generated insights on where to focus next.
🔐 Secure Auth — JWT-based authentication with role-aware access control across web and mobile.
- Python 3.9+, Node.js 18+, Docker & Docker Compose
# 1. Clone
git clone https://github.com/johnmoses/ultra-learning.git
cd ultra-learning
# 2. Configure environment
cp .env.dev .env
# Edit .env: LLM API keys, DB credentials
# 3. Start backend + database
docker-compose up -d --build
# 4. Run migrations
cd api && pip install -r requirements.txt && flask db upgrade && cd ..
# 5. Start web frontend
cd web && npm install && npm run dev # → http://localhost:3000
# 6. Start mobile
cd mobile && npm install && npx react-native run-iosAgent Orchestration — Built with LangChain and Anthropic's Model Context Protocol (MCP), agents query live application state (user progress, document store, flashcard performance) without hallucinating context.
RAG Pipeline — Documents are chunked with overlap-aware splitting, embedded via sentence-transformers, and stored in Milvus. Retrieval uses semantic search with a re-ranking step before LLM context injection.
Streaming — All AI responses stream to the client via Server-Sent Events (SSE), giving sub-second time-to-first-token.
⚡ Complete proof of concept — fully functional locally. All core features implemented. Deployment to cloud (GCP Cloud Run + Vercel) planned.
Built by John Moses — co-founding engineer at ShopStack360 and independent AI consultant through AXIIS Microsystems.
MIT — see LICENSE for details.