0
stars
88
commits
Python
primary language
Jun 15, 2026
updated
README
HudlScanner
Extract American football plays from playbook images using vision-language models. Upload playbook pages (PNG/PDF), pick a position, and get a structured, editable play table you can export to Excel.
Stack
- Frontend/API: SvelteKit (Svelte 5, TypeScript, Tailwind CSS), deployed on Vercel
- Extraction models: Local dots.mocr (Docker/vLLM) with cloud fallbacks (Anthropic Claude → Moonshot Kimi → DeepSeek)
- ML pipeline: Python scripts for building training data and fine-tuning position-specific models
Getting started
npm install
cp .env.example .env.local # then fill in your API keys
npm run dev
The app runs at http://localhost:5173 (or 5174). At least one extraction backend must be configured in .env.local:
| Variable | Purpose |
|---|---|
USE_LOCAL_MOCR | true to use the local Docker model first |
LOCAL_MOCR_URL | Local vLLM server URL (default :8000) |
ANTHROPIC_API_KEY | Claude (primary cloud fallback) |
MOONSHOT_API_KEY | Moonshot Kimi (secondary fallback) |
DEEPSEEK_API_KEY | DeepSeek (tertiary fallback) |
ALLOWED_ORIGINS | Comma-separated CORS allowlist for the API |
Commands
npm run dev # dev server
npm run build # production build
npm run preview # preview production build
npm run check # svelte-check (type checking)
npm run lint # prettier + eslint
npm run format # prettier --write
npm run test:unit # vitest unit tests
npm run test:e2e # playwright e2e tests (run `npx playwright install` once first)
npm run test # unit + e2e
Project layout
| Path | Purpose |
|---|---|
src/routes/ | Pages (/, /upload, /editor, marketing pages) |
src/routes/api/extract-plays/ | Extraction API endpoint |
src/lib/components/ | UI components |
src/lib/prompt.js | Vision prompt builder + playbook vocabulary |
src/lib/server/extraction/ | Provider chain, response parsing, request schema |
src/lib/server/route-tree-data.js | Base64 route tree legend (regenerate: python scripts/gen_route_tree.py) |
scripts/ | Python ML pipeline (training data, fine-tuning, annotation) |
training/ | Training images and JSONL datasets |
docker/ | Local dots.mocr inference server (vLLM) |
hf-space/ | Hugging Face Spaces inference server |
docs/ | Setup guides, design docs, mockups |
Docs
- docs/QUICK_START.md — quick start guide
- docs/BACKEND_SETUP_GUIDE.md — backend setup
- docs/LOCAL_AI_SETUP.md — local model via Docker
- CLAUDE.md — full project reference (training pipeline, concept rules, hardware)
Contributors
LordLP90000
88 commits
Similar repos
Languages
Python
41.2%
Svelte
33.8%
HTML
12.8%
TypeScript
8.1%
JavaScript
3.1%
0
stars
88
commits
Python
primary language
Jun 15, 2026
updated
README
HudlScanner
Extract American football plays from playbook images using vision-language models. Upload playbook pages (PNG/PDF), pick a position, and get a structured, editable play table you can export to Excel.
Stack
- Frontend/API: SvelteKit (Svelte 5, TypeScript, Tailwind CSS), deployed on Vercel
- Extraction models: Local dots.mocr (Docker/vLLM) with cloud fallbacks (Anthropic Claude → Moonshot Kimi → DeepSeek)
- ML pipeline: Python scripts for building training data and fine-tuning position-specific models
Getting started
npm install
cp .env.example .env.local # then fill in your API keys
npm run dev
The app runs at http://localhost:5173 (or 5174). At least one extraction backend must be configured in .env.local:
| Variable | Purpose |
|---|---|
USE_LOCAL_MOCR | true to use the local Docker model first |
LOCAL_MOCR_URL | Local vLLM server URL (default :8000) |
ANTHROPIC_API_KEY | Claude (primary cloud fallback) |
MOONSHOT_API_KEY | Moonshot Kimi (secondary fallback) |
DEEPSEEK_API_KEY | DeepSeek (tertiary fallback) |
ALLOWED_ORIGINS | Comma-separated CORS allowlist for the API |
Commands
npm run dev # dev server
npm run build # production build
npm run preview # preview production build
npm run check # svelte-check (type checking)
npm run lint # prettier + eslint
npm run format # prettier --write
npm run test:unit # vitest unit tests
npm run test:e2e # playwright e2e tests (run `npx playwright install` once first)
npm run test # unit + e2e
Project layout
| Path | Purpose |
|---|---|
src/routes/ | Pages (/, /upload, /editor, marketing pages) |
src/routes/api/extract-plays/ | Extraction API endpoint |
src/lib/components/ | UI components |
src/lib/prompt.js | Vision prompt builder + playbook vocabulary |
src/lib/server/extraction/ | Provider chain, response parsing, request schema |
src/lib/server/route-tree-data.js | Base64 route tree legend (regenerate: python scripts/gen_route_tree.py) |
scripts/ | Python ML pipeline (training data, fine-tuning, annotation) |
training/ | Training images and JSONL datasets |
docker/ | Local dots.mocr inference server (vLLM) |
hf-space/ | Hugging Face Spaces inference server |
docs/ | Setup guides, design docs, mockups |
Docs
- docs/QUICK_START.md — quick start guide
- docs/BACKEND_SETUP_GUIDE.md — backend setup
- docs/LOCAL_AI_SETUP.md — local model via Docker
- CLAUDE.md — full project reference (training pipeline, concept rules, hardware)
Similar repos
Contributors
LordLP90000
88 commits
Languages
Python
41.2%
Svelte
33.8%
HTML
12.8%
TypeScript
8.1%
JavaScript
3.1%