Hack Win #24 — The General Partnership Hackathon
Dropped into an on-the-spot challenge with zero prep, solved it in record time: under 6 hours. One of the most creatively designed and high-impact problem statements I’ve tackled yet.
The Challenge
The General Partnership Hackathon threw a curveball — no advance notice on the problem statement. You show up, hear the challenge, and build. Pure problem-solving under pressure.
The task: build a real-time sports analytics platform that could ingest live game data, generate tactical insights, and present actionable strategies to coaching staff — all within a single session.
What We Built
Super Analytics — a real-time sports analytics and strategy platform that processes live game events, applies multi-agent analysis, and delivers coaching recommendations in real-time.
Architecture
The system runs on a three-layer pipeline:
- Data Ingestion Layer — Streams live game events (player positions, ball movement, fouls, substitutions) through a WebSocket pipeline into a normalized event store
- Analysis Engine — A multi-agent system where specialized agents handle different aspects:
- Pattern Agent — Detects offensive and defensive formations using spatial clustering algorithms
- Momentum Agent — Tracks game momentum shifts by analyzing scoring runs, turnovers, and possession changes over sliding time windows
- Matchup Agent — Evaluates player-vs-player effectiveness using historical and real-time performance metrics
- Strategy Layer — Synthesizes agent outputs into coaching recommendations: lineup changes, play suggestions, and timeout triggers based on momentum analysis
Technical Stack
- Backend — Python FastAPI for real-time WebSocket connections and REST endpoints
- Agent Orchestration — LangGraph for coordinating analysis agents with shared state
- Data Processing — Pandas and NumPy for real-time statistical computations
- Frontend — React with D3.js for live court/field visualizations and dynamic charts
- LLM Integration — GPT-4 for natural language strategy summaries and coaching narrative generation
Key Features
- Live Dashboard — Real-time visualization of player movements, shot charts, and possession flow
- AI-Generated Play-by-Play — Natural language summaries of key game moments with strategic context
- Predictive Alerts — Proactive notifications when momentum shifts suggest a timeout or substitution
- Historical Comparison — Overlays current game patterns against historical matchup data for trend analysis
The Result
Winner! Built and presented in under 6 hours with zero prior preparation.
What Made It Work
The key to winning wasn’t just speed — it was architectural discipline under pressure. Instead of building a monolithic app, we designed independent agents that could be developed and tested in parallel, then composed into the final system. This let us divide work cleanly and integrate without merge conflicts.
The multi-agent approach also made the demo compelling — showing how different analytical perspectives combine to produce insights that no single model could generate alone.
Takeaway
Big thanks to The General Partnership for hosting one of the most creatively designed hackathons I’ve encountered — real-world complexity with clear impact.