MeatLink B2B Meat Commerce Platform

The problem

Restaurants, hotels, caterers, and butcher shops in Indonesia often buy meat through scattered WhatsApp chats and supplier price lists, with little visibility into stock, grades, or fair per-kilogram prices. Suppliers lack a simple channel to reach business buyers, and beef-market intelligence is spread across FAOSTAT, USDA, and Indonesian government sources.

What I built

Built meatlink.id with a partner as a full B2B commerce platform: catalog and search with transparent per-kilogram pricing, multi-item RFQs with reference tracking, cart and checkout with Midtrans payments, orders and invoices, and admin and vendor consoles for inventory, AI-assisted spreadsheet import, payment reconciliation, and reporting. It runs TanStack Start on Cloudflare Workers with Supabase Postgres and exposes an OAuth 2.1-protected MCP server, so AI agents can check stock, place and look up orders, and draft market insights. A separate analytical engine runs a Codex research agent that turns public market data into bilingual EN/ID insight drafts without ever holding database credentials.

Proof

1. 01 Row-Level Security policies per role (anonymous, buyer organization, vendor organization, platform admin), checked against a written RLS test matrix after every policy change.
2. 02 Hardened payment lifecycle: Midtrans webhook handling, 24-hour payment expiry aligned with order cancellation, a scheduled sync every 30 minutes, and an admin-only daily reconciliation that flags mismatches instead of force-settling orders.
3. 03 MCP endpoint at meatlink.id/mcp with tools for stock, orders, market data, and insight drafts; server-side gates enforce evidence freshness, block internal marketplace data, and require complete English and Indonesian text.
4. 04 AI writes are draft-only: every agent-generated insight is reviewed, edited, and published by a human from the admin console.
5. 05 Built with an AI-native workflow (Lovable, Codex) while owning the architecture, data model, security policies, integrations, Vitest test suites, k6 load-test scenarios, and operational runbook.