Multi-Service AI Inspection Pipeline: Live on iOS and Android

Inspectors were spending their evenings turning field notes and photos into client-ready reports. The build looks simple until you count the systems involved: capture in the field, extract meaning from messy input, generate a heavy PDF, notify the user, and give an admin visibility over all of it — with none of them allowed to fail silently, and a user waiting at the end.

We architected and shipped five services. A mobile app handles capture, offline tolerance, billing and push. A Python backend runs a dual-provider LLM pipeline that extracts structured findings from inspection input. A dedicated PDF service handles image compression and report rendering. A background worker polls the job table every two seconds and runs a ten-step pipeline: analysis, mini reports, final report, credit deduction, image compression, PDF generation, storage upload, database update, then push notification. An admin dashboard sits over the top. Error tracking runs on every service.