What this is
A multi-tenant inventory and commerce platform we built for market owners who manage stock across one or many stores. The painful first step — turning the supplier documents that arrive in the back office (invoices, packing lists, receipts, restocking sheets) into structured inventory records — is handled by a document-AI ingestion path with a human-review gate before stock counts move.
The product is mobile-first because the operators it serves run their businesses from a phone.
Why we built it
Inventory in SMB and emerging-market retail is dominated by paper and ad-hoc digital documents. Generic point-of-sale systems assume a clean, pre-existing SKU catalog and a reliable barcode scanner. Real operators don’t have either. We wanted to validate that document AI plus a careful product UX could give a small market owner the inventory clarity that a large retailer takes for granted — without forcing them into a system designed for someone else.
What it does
- Multi-tenant SaaS — each store’s data is isolated, with shared platform infrastructure.
- Ingests inventory documents (PDFs, photos, scans) and turns them into structured inventory records via OCR and vision-AI parsing.
- Keeps a human in the loop: parsed documents land in a review queue with the source image and extracted fields before stock counts move.
- Tracks stock movement with manual entry as the current default; designed to extend to point-of-sale, online store, and multi-store management in later phases.
- Mobile-first interface tuned for operators who manage their business from a phone.
Built with
| Layer | Stack |
|---|---|
| Frontend | React + TypeScript + Vite + Tailwind CSS, deployed on Vercel |
| Backend | Express.js (Node.js) |
| Database | PostgreSQL — managed in production, Docker locally |
| AI document parsing | Commercial vision-language API |
| OCR fallback | Open-source OCR engine |
| Reproducibility | Docker Compose (clean clone → running stack) |
What this means for you
If you operate a business where the source-of-truth records arrive as documents — paper, photos, PDFs, screenshots — and the cost of getting them into a structured system is currently borne by someone typing them in by hand, we can:
- Build a document-OCR-to-structured-data system tuned to your actual document corpus.
- Keep a human-review gate in the workflow where the cost of being wrong is high.
- Deliver as a multi-tenant SaaS or single-tenant self-hosted system, mobile-first or desktop-first, depending on how your users actually work.
Want to talk about a similar problem in your business? Contact us.