Demo build with fictional stores, sample prices and sample model outputs; written in Python and structured like BuildShip workflows.
I made a working reference backend for a WhatsApp grocery price agent aimed at the Saudi and Gulf market, written in Python and structured like BuildShip workflows, with sample model outputs. A 3:00 AM Riyadh cron scrapes five fictional stores; a cleaning step turns 44 raw rows into offers and a validation node keeps 40 and rejects 4. The handler reads text, list photos (Vision-to-JSON) and voice notes, compares single units with bulk cartons and replies in dialect-friendly Arabic in at most 1.2 ms p95 of backend time on fixtures; when a model call stalls, a holding reply goes out inside 5 seconds. A signed payment webhook activates subscriptions. 50 of 50 tests pass.
Overview and the 03:00 scrape workflow, with the run and the rejected rows
A photo of a written shopping list, its Vision-to-JSON output (a sample) and the dialect-friendly Arabic reply
Lowest price per store, single unit vs carton, and the measured latency
Payment webhook workflow, cases and signature check, and the adapter for each service
0 likes · 1 view · Oct 3, 2026
A reference backend for a WhatsApp grocery assistant, with sample model outputs: a 3:00 AM scrape with a cleaning step, a handler that reads text, list photos and voice notes and replies in dialect-friendly Arabic, and a signed payment webhook. Fictional stores and prices.
