AI-powered fight analysis platform that processes combat sports footage through a multi-agent system. Upload fight video, get structured breakdown of technique, strategy, scoring, and improvement recommendations — powered by LangGraph agent orchestration.
┌──────────────────────────────────────────────────────────┐
│ Next.js 15 Frontend (React 19 + Framer Motion) │
│ ┌──────────┐ ┌──────────┐ ┌───────────────────────┐ │
│ │ Upload │ │ Live │ │ Analysis Dashboard │ │
│ │ Panel │ │ Feed │ │ (Charts + Timeline) │ │
│ └────┬─────┘ └────┬─────┘ └───────────┬───────────┘ │
│ └──────────────┴────────────────────┘ │
│ │ REST API │
├───────────────────────┼──────────────────────────────────┤
│ FastAPI Backend │ │
│ ▼ │
│ ┌─────────────────────────────────────────────────┐ │
│ │ LangGraph StateGraph │ │
│ │ │ │
│ │ ┌───────────┐ ┌────────────┐ │ │
│ │ │ Fight │───▶│ Validator │ │ │
│ │ │ Analyst │ │ Agent │ │ │
│ │ │ Agent │ └─────┬──────┘ │ │
│ │ └───────────┘ │ │ │
│ │ │ ┌────▼──────┐ │ │
│ │ │ │ Scoring │ │ │
│ │ └───────────▶│ Engine │ │ │
│ │ └───────────┘ │ │
│ └─────────────────────────────────────────────────┘ │
│ │ │
│ ┌────────┴────────┐ │
│ │ PostgreSQL │ │
│ │ (Supabase) │ │
│ └─────────────────┘ │
└──────────────────────────────────────────────────────────┘
1. Video Ingestion — Upload fight footage via the frontend. The backend extracts metadata, timestamps, and prepares analysis context.
2. Multi-Agent Analysis — A LangGraph StateGraph orchestrates specialized agents:
- Fight Analyst Agent — Contract-driven agent that breaks down technique exchanges, identifies patterns, and assesses fighter strategies
- Validator Agent — Cross-references analysis claims against video evidence and known technique databases
- Scoring Engine — Applies judging criteria to generate round-by-round or segment-based scoring
3. Structured Output — Results are assembled into a comprehensive fight report with timeline annotations, technique classifications, strategic assessments, and actionable coaching feedback.
| Layer | Technology |
|---|---|
| Frontend | Next.js 15 (App Router), React 19, TypeScript, Framer Motion, TailwindCSS |
| Backend | FastAPI, Python 3.11+, Pydantic V2 |
| Agent Framework | LangGraph (StateGraph with conditional routing) |
| Agent Behavior | JSON contract-based prompt architecture |
| Database | PostgreSQL via Supabase (pgvector for embeddings) |
| Auth | Supabase Auth with cookie-based sessions |
| Testing | Pytest (backend), Jest (frontend) |
| Infrastructure | Docker, Docker Compose |
- Python 3.11+
- Node.js 18+
- Supabase account (free tier works)
- OpenAI API key
cd fight-analyst-backend
python -m venv .venv
.venv/Scripts/activate # Windows
source .venv/bin/activate # Mac/Linux
pip install -r requirements.txt
cp .env.example .env # Add your API keys
python -m uvicorn app.main:app --reload --host 0.0.0.0 --port 8001cd fight-analyst-frontend
npm install
cp .env.example .env.local # Set NEXT_PUBLIC_BACKEND_URL
npm run dev # http://localhost:3000docker-compose up --buildfight-analyst-backend/
├── app/
│ ├── agents/ # LangGraph agent definitions
│ │ ├── contracts/ # JSON behavior contracts
│ │ ├── fight_analyst.py
│ │ ├── validator.py
│ │ └── base.py
│ ├── graph/ # LangGraph StateGraph wiring
│ ├── api/ # FastAPI route handlers
│ ├── services/ # Business logic layer
│ ├── models/ # Pydantic schemas
│ ├── memory/ # Session and persistent memory
│ ├── middleware/ # Auth, CORS, rate limiting
│ └── tests/ # Pytest test suite
├── migrations/ # Database migrations
├── Dockerfile
└── requirements.txt
fight-analyst-frontend/
├── app/ # Next.js 15 App Router pages
├── middleware.ts # Auth middleware
├── package.json
└── tsconfig.json
Contract-Driven Agents — Agent behavior is defined in JSON specification files, not hardcoded in Python. This enables modifying agent personality, scope, and response patterns without touching application code.
LangGraph over raw chains — StateGraph provides deterministic routing, state persistence, and conditional branching that raw LangChain chains can't offer for multi-step analysis workflows.
Validator as separate agent — Rather than self-validation, a dedicated validator agent catches hallucinations and unsupported claims by cross-referencing against the original video context.
This project combines 28+ years of experience coaching professional combat athletes with production AI agent engineering. The analysis models are grounded in real fight coaching methodology — not generic sports analytics.
MIT
Pete Davidsmeier — AI Agent Architect, CTO at YBRYX Business Systems
