IScored is an AI-powered sports streaming and management platform. It ingests live camera feeds from the field, streams them with low latency, and uses computer vision to automatically generate highlights and game analytics.
Highlights
-
Low-latency streaming architecture — Architected a video ingestion pipeline routing multiple IP camera streams (RTSP) through secure VPN tunnels to a centralized server, using MediaMTX to transmux feeds into HLS for real-time playback.
-
Automated AI highlight generation — Built a computer-vision pipeline with YOLOv8 that analyzes full 45-minute match recordings, identifies key events (goals, fouls, shots), and clips them into concise 5-minute highlight reels.
-
Scalable GPU inference — Decoupled heavy ML processing onto Modal serverless GPUs, scaling player tracking and ball-possession analysis on demand without impacting core server performance.
-
Video management infrastructure — Integrated Mux for adaptive-bitrate streaming and AWS S3 for storage, orchestrating the full lifecycle (upload, encoding, archiving) through an event-driven webhook architecture.
-
Computer-vision analytics — Implemented object-detection algorithms to track player movement and possession percentages, turning raw video into actionable game insights.
Tech Stack
Node.js · YOLOv8 · OpenCV · MediaMTX · Modal (Serverless GPU) · Mux · PostgreSQL · AWS S3