📄 DocFind — Global Lost-Document Finder

A service that helps reunite people with lost documents (passports, IDs, driver's licenses) using OCR + owner verification over WhatsApp.

🔗 Live: doc.caim.dev

🔒 This is a showcase. Source code is in a private repository. This page documents the architecture and key engineering wins.

---

💡 How it works

1. Someone who found a document uploads photos + a short silent video (camera only, no gallery).
2. The system runs OCR to extract the document's identifying fields.
3. The owner claims it and proves ownership via a WhatsApp OTP.
4. On a verified match, the two are connected.

Country-aware document types and field rules (RU passport / СНИЛС / ИНН, US SSN, India PAN / Aadhaar, Kyrgyz AC/AN/ID prefixes, …) and a full admin panel.

---

⚡ Key engineering win — OCR from ~180s to ~1s

PaddleOCR was installed but misconfigured and painfully slow. After profiling, a single fix (enable_mkldnn) plus pipeline tuning brought a document scan from ~180 seconds down to ~1 second — a ~180× speedup — with Tesseract.js + Mindee as fallbacks.

---

🏗️ Architecture

Client (camera capture) ──► Fastify API ──► PostgreSQL (Prisma)
                                  │
                                  └──► OCR sidecar (FastAPI + PaddleOCR 3.5)
                                  └──► WhatsApp OTP verification

---

🛠️ Tech stack

API Fastify · Prisma · PostgreSQL OCR FastAPI · PaddleOCR 3.5 · Tesseract.js · Mindee (fallback) Verification WhatsApp OTP Infra Linux · Nginx · PM2

---

Built by Muratbek Omurzakov.