I built a document extraction API that runs on regex (no LLM needed)

A developer built Invoice to JSON Extractor, a public API that parses invoices, receipts, resumes, and bank statements into structured JSON using a regex heuristic engine rather than an LLM. The base path handles roughly 70% of typical invoices in under 50ms at zero ongoing cost, escalating to DeepSeek or Groq only when the regex extractor returns a low confidence score, so about 95% of requests never hit an LLM. The service, built on FastAPI and Pydantic and billed through RapidAPI, offers a free tier of 500 calls per month with paid tiers starting at $9/month for 2,000 calls.

I kept rewriting invoice parsing code on every freelance project. Same regex, same edge cases, same pain. So I made it a public API instead. It's called Invoice to JSON Extractor . Four endpoints, one auth, text in, structured JSON out. The interesting part: the base path doesn't use an LLM. It's a regex heuristic engine that handles ~70% of typical invoices in under 50ms, for zero ongoing cost. | Endpoint | Input | Output | |---|---|---| | POST /v1/invoice/extract | Invoice text | number, dates, vendor, VAT, totals, line items | | POST /v1/receipt/extract | Receipt text | store, date, items, total, payment method | | POST /v1/resume/extract | CV/resume text | name, email, phone, skills, experience | | POST /v1/bank-statement/extract | Bank statement text | account, transactions, running balance | Bilingual out of the box English + Vietnamese . When I started, I assumed I'd need an LLM for this. But most invoices follow one of maybe 15 common layouts. Regex handles these in 40-60ms with zero API calls. The LLM is a fallback , not the primary engine. If the regex extractor returns a low confidence score, the API escalates to DeepSeek or Groq for that specific request. Result: ~95% of requests never hit an LLM. Average cost per request rounds to zero. Input: INVOICE Invoice : INV-2024-001 Date: 15/03/2024 Vendor: Acme Supplies Ltd. VAT: GB123456789 Bill To: Widget Corp Widget A 10 5.00 50.00 Widget B 3 20.00 60.00 Subtotal: 110.00 VAT 20%: 22.00 Total: 132.00 GBP Output: { "success": true, "invoice": { "invoice number": "INV-2024-001", "invoice date": "2024-03-15", "currency": "GBP", "vendor name": "Acme Supplies Ltd.", "vendor tax id": "GB123456789", "customer name": "Widget Corp", "subtotal": 110.0, "tax amount": 22.0, "total": 132.0, "line items": {"description": "Widget A", "quantity": 10.0, "unit price": 5.0, "amount": 50.0}, {"description": "Widget B", "quantity": 3.0, "unit price": 20.0, "amount": 60.0} , "confidence": 0.9 }, "model": "heuristic:v1", "processing ms": 42 } curl -X POST "https://invoice-to-json-extractor1.p.rapidapi.com/v1/invoice/extract" \ -H "content-type: application/json" \ -H "X-RapidAPI-Key: YOUR KEY" \ -H "X-RapidAPI-Host: invoice-to-json-extractor1.p.rapidapi.com" \ -d '{"content":"Invoice : INV-1\nDate: 2024-01-15\nTotal: 132.00 USD"}' Get a free key here: Invoice to JSON Extractor https://rapidapi.com/tuyentn23/api/invoice-to-json-extractor1 Free tier: 500 calls/month. Paid tiers start at $9/month for 2,000 calls. If you build anything that processes invoices or receipts, I'd love feedback on what fields you actually need. The API is small and I can iterate fast. Stack: FastAPI + Pydantic, deployed on Render free tier, billing via RapidAPI. Total monthly cost so far: $0.