🧠 Multi-Agent Competitive Programming Coach

An AI-powered coaching platform that uses a team of specialized agents to teach competitive programming through Socratic dialogue — never giving direct answers, always guiding through questions.

Architecture

┌──────────────────────────────────────────────────────┐
│                    Frontend (Next.js)                │
│              Real-time SSE streaming UI              │
└────────────────────────┬─────────────────────────────┘
                         │ REST + SSE
┌────────────────────────▼─────────────────────────────┐
│                  FastAPI Backend                     │
│  ┌─────────────┐  ┌──────────┐  ┌────────────────┐   │
│  │  Auth (JWT) │  │ Sessions │  │  SSE Streaming │   │
│  └─────────────┘  └──────────┘  └────────┬───────┘   │
│                                          │           │
│  ┌───────────────────────────────────────▼────────┐  │
│  │            LangGraph Supervisor                │  │
│  │  ┌──────────────────────────────────────────┐  │  │
│  │  │  Problem   │ Teaching │ Algorithm │ Code │  │  │
│  │  │  Analyzer  │  Agent   │  Expert   │Review│  │  │
│  │  ├──────────────────────────────────────────┤  │  │
│  │  │ Complexity │  Test    │ Learning  │      │  │  │
│  │  │ Analyzer   │  Cases   │  Memory   │      │  │  │
│  │  └──────────────────────────────────────────┘  │  │
│  └────────────────────────────────────────────────┘  │
│                                                      │
│  ┌────────────┐  ┌──────────┐  ┌──────────────────┐  │
│  │ PostgreSQL │  │  Qdrant  │  │      Redis       │  │
│  │  (state)   │  │ (vectors)│  │    (caching)     │  │
│  └────────────┘  └──────────┘  └──────────────────┘  │
└──────────────────────────────────────────────────────┘

Tech Stack

Layer	Technology
Frontend	Next.js 14, TypeScript, Tailwind
Backend	FastAPI (async), Python 3.12+
Agent Framework	LangGraph + LangChain
LLM	OpenAI GPT-4o
Database	PostgreSQL 16 (async via asyncpg)
Vector DB	Qdrant
Cache	Redis 7
Auth	JWT (python-jose + passlib)

Agent System

Agent	Role
Supervisor	Routes queries to the right specialist agent
Problem Analyzer	Parses problems, estimates difficulty, identifies categories
Teaching Agent	Core Socratic coach — guides with questions, not answers
Algorithm Expert	Explains algorithms, compares approaches pedagogically
Complexity Analyzer	Analyzes time/space complexity, validates student analysis
Test Case Generator	Creates edge cases, explains why each matters
Code Review	Reviews code WITHOUT rewriting — only explains issues
Learning Memory	Tracks proficiency, identifies patterns and weaknesses

Two Modes

• Learning Mode — Full Socratic coaching with progressive hints (levels 0–5)
• Contest Mode — Timer, no hints, only clarification questions allowed

Quick Start

Prerequisites

• Docker & Docker Compose
• OpenAI API key

Setup

# 1. Clone and configure
cp .env.example .env
# Edit .env with your OPENAI_API_KEY

# 2. Start all services
make dev

# 3. Access
# Backend API:  http://localhost:8000
# API Docs:     http://localhost:8000/docs
# Frontend:     http://localhost:3000

Local Development (without Docker)

# Start infrastructure
docker compose up -d postgres redis qdrant

# Install backend deps
cd backend
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt

# Run backend
uvicorn app.main:app --reload --port 8000

API Endpoints

Method	Path	Description
POST	/api/auth/register	Register new user
POST	/api/auth/login	Login, get JWT
POST	/api/sessions/	Create coaching session
GET	/api/sessions/	List sessions
GET	/api/sessions/{id}	Session details + messages
POST	/api/sessions/{id}/chat/stream	SSE streaming chat
POST	/api/sessions/{id}/submit-code	Submit code for review
PATCH	/api/sessions/{id}	Update session status
GET	/api/profile/	User learning profile
GET	/api/profile/progress	Progress over time
GET	/api/problems/	List problems
GET	/health	Health check

License

MIT