Harvey Xia
Back to projects

PandaMart PMS

Inventory, procurement & AI assistanceInteractive demo

From a supplier invoice to stock a small team can trust.

PandaMart operations dashboard with pending invoices, inventory risks and purchasing trends from seeded demo data
Operations dashboard · Public demo with generated business data

An inventory and procurement platform connects invoice review, stock, expiry dates and supplier balances. A local-first assistant answers operational questions; an LLM integration is optional.

Status
Public demo
AI in the product
Rules-based assistant; optional LLM; browser OCR
Evidence
Live demo and source code
Limits
Seeded demo data, not a customer deployment

The business question

How can a small retail team turn incoming invoices into reliable stock records without treating extracted text as automatically correct?

The workflow

Extract invoice fields, review line items, then confirm the transaction. Bring low-stock items, expiring batches and payables into the same operational view. Keep a rules-based assistant available when no LLM is configured.

What the demo proves

The public demo runs with a browser-side mock backend and seeded records. Image and PDF OCR use browser-side Tesseract; the assistant queries the demo data through local rules. It is an inspectable workflow, not evidence of a production customer deployment.

Choose a demo account on the login screen. Records stay in your browser; you can reset the sample data.

Dated implementation evidence

  1. Keep a local-first assistant available alongside optional LLM integration.

    View commit · 98ef387
  2. Run invoice OCR in the public browser demo with Tesseract.js.

    View commit · 98b94e1

Verification and limits

The commits establish implemented workflows, not OCR accuracy on a customer invoice collection. The public demo uses seeded data.

AI collaboration: attribution boundary

The records below show implementation and review outcomes. They do not record a complete prompt history or identify human versus AI authorship of individual changes. Commit history alone cannot establish that split; no percentage of authorship is claimed.