Gelora Football Video Analysis
Web app for football coaches and analysts to tag match events, cut clips, share playlists, and run a real offline person-detection trial.
Description
A local-first Next.js + Prisma app for football video analysis. Five-role auth (Admin, Analyst, Head Coach, Assistant Coach, Player) with a demo role-switcher, a full video analyzer (timeline, hotkey tagging, manual clip creation, telestration drawing with player/cone markers and manual distance/speed tools), a Sportscode-style Scout coding panel, playlist editor with reel export, and a keyword-based Search screen that is explicitly labeled as keyword search, not real semantic/NLU search. The analyzer's People tab runs a genuine offline YOLOX-S person-detection trial via OpenCV DNN — separate from the Review tab's AI event suggestions, which are clearly and honestly simulated (every simulated result is tagged gelora-sim- in the database).
What Already Works
- Auth + RBAC across 5 roles with a demo role-switcher
- Video library, match detail, and analyzer with hotkey tagging and manual clip creation
- Telestration drawing: shapes, player/cone widget markers, undo/redo, manual distance/speed measurement
- Sportscode-style Scout coding panel with per-player, per-tag timeline rows
- Playlist editor with reordering, timestamp-anchored comments, and reel export (ffmpeg concat)
- Real offline YOLOX-S person-detection trial tab, publishing a browser-safe annotated clip
- Keyword Search screen (explicitly labeled — not semantic/NLU search) with click-to-seek results
- Review tab AI event suggestions — clearly labeled simulated, never auto-confirmed
What's Included
- Full source code
- README and setup guide
- InMyDraft Commercial License
- AI-ready docs (PRD, technical spec, database schema, ML pipeline, deployment notes)
- Prisma seed data
How to Run
- Unzip the source and run npm install.
- Run npx prisma generate, npm run db:push, npm run db:seed.
- Run npm run dev and open http://localhost:3000.
- For the People tab trial: create a Python venv, pip install -r requirements-vision.txt, and make sure FFmpeg is on PATH.
Continuation Notes
- Every simulated AI result is tagged gelora-sim- in the database so it's never mistaken for a real detector result — worth keeping that convention if you extend it.
Known Limitations
- Semantic Search is keyword-only over confirmed tags/notes/titles — real semantic search would need a paid VLM/embeddings API, deliberately not included
- Review-tab AI event suggestions are simulated, not a real detector — no GPU worker or ByteTrack pipeline is included
- Person-detection trial counts people-visible-per-frame, not unique player identities
- Team assignment, ball tracking, and derived sprint/possession metrics are not built
- The person-detection trial needs a local Python environment and FFmpeg on PATH in addition to Node
Related Drafts
BatuWorkout Class Booking
Workout class booking draft with public schedule pages, booking/member/admin screens, and manual-payment MVP direction.
Personal Finance Tracker
Half-built but running web app draft for tracking personal income, expenses, and simple cashflow habits.
