वही · your finance toolkit — turn invoice PDFs into Excel, fuzzy-match any two lists. ₹0 / $0 per document.

Live demo: vahi-lemon.vercel.app Backend: vishwas51-vahi-api.hf.space Stress-tested accuracy: 93.1% invoice OCR · 96.4% fuzzy match

Built for finance teams hitting two daily walls:

1. Invoice OCR — drop digital + scanned PDFs, get one merged Excel with custom column headers.
2. Fuzzy Lookup — match two lists (bank statement → vendor master, etc.) with smart business-name normalization.

Architecture: Next.js frontend (built with Gemini) + FastAPI backend (Python). Both deploy free, both global.

.
├── api.py                       # FastAPI backend (4 endpoints)
├── modules/                     # Pure-Python invoice/fuzzy logic
│   ├── pdf_utils.py
│   ├── invoice_ocr.py
│   ├── fuzzy_lookup.py
│   ├── excel_writer.py
│   ├── column_utils.py
│   └── schema.py
├── requirements.txt             # Python deps
├── packages.txt                 # System deps for HF Spaces (tesseract, poppler)
├── Dockerfile                   # Backend container
├── vahi-frontend/               # Next.js + Tailwind + shadcn/ui (Gemini-generated)
└── tests/                       # Stress test data + benchmark runner

Terminal 1 — backend:

brew install tesseract poppler        # macOS; on Linux: apt-get install tesseract-ocr poppler-utils
python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
uvicorn api:app --reload --port 8000

Sanity-check: curl http://localhost:8000/api/health → {"ok":true,...}.

Terminal 2 — frontend:

cd vahi-frontend
cp .env.example .env.local
npm install
npm run dev

Open http://localhost:3000.

The recommended stack:

1. Sign up free at https://huggingface.co/join.
2. Create a new Space at https://huggingface.co/new-space:
   • Name: vahi-api
   • SDK: Docker
   • Hardware: CPU basic — Free
   • Visibility: Public
3. Get a write token at https://huggingface.co/settings/tokens.
4. Push this repo to that Space:

   cd "/Users/vishwaspandey/OCR cum FUZZY LOOKUP"
   git remote add space https://huggingface.co/spaces/<your-username>/vahi-api
   git push space main      # paste the write token when prompted

5. HF reads the Dockerfile at the repo root, builds the image, and runs it on port 7860.
6. After ~5 min build: https://<your-username>-vahi-api.hf.space/api/health returns {"ok":true}.

The backend listens on $PORT (default 7860 on HF). If your hosted URL is https://abc-vahi-api.hf.space, your API base is https://abc-vahi-api.hf.space/api.

1. Push the repo to GitHub (so Vercel can read it):

   gh repo create vahi --public --source=. --push

2. Go to https://vercel.com/new, import the GitHub repo.
3. Root Directory: vahi-frontend (important — Vercel needs to point at the Next.js app).
4. Environment Variables: add one:

   NEXT_PUBLIC_API_BASE = https://<your-username>-vahi-api.hf.space/api
   

5. Click Deploy. ~2 minutes.
6. Vercel gives you https://vahi-<random>.vercel.app. Anyone in the world can use it.

Edit the HF Space's Settings → Variables and add:

VAHI_ALLOWED_ORIGINS=https://vahi-<random>.vercel.app,https://*.vercel.app,http://localhost:3000

Restart the Space. Now only your frontend (and localhost dev) can hit the API.

For "everyone in the world, free", HF Spaces wins. Period.

OpenAPI / Swagger UI is auto-generated at /docs.

Run python tests/run_stress.py from the repo root.

Performance: 25 ms per digital invoice, ~7 s per scanned, 0.4 ms per fuzzy row.

• The backend never writes uploaded PDFs to disk.
• Job results are kept in-memory for 30 min, then garbage-collected.
• For sensitive invoices, self-host the Docker container behind your VPN.

MIT.