An AI field-service platform for Danish trade companies

WorcX serves Danish trade companies from a mobile app used on-site — but the on-the-ground reality made several features genuinely hard to build:

• Tradespeople rarely write structured inspection notes, and freeform text can’t drive automated material ordering.
• Turning a plain-language job description into a costed, multi-supplier bill of materials was manual, slow and error-prone.
• Signup needed live Danish company verification — but the official registry is EU-restricted and strictly rate-limited.
• 900+ UI strings were hardcoded, so every wording tweak meant a full app release.
• The whole production stack needed to move to leaner infrastructure without losing a byte.

Software Alliance built the platform as a NestJS microservices monorepo — one API gateway fronting six domain services over TCP RPC, with PostgreSQL / Prisma, a Next.js admin panel and Stripe billing — and shipped five features end-to-end:

• AI-assisted inspection notes. A 7-question, admin-managed tap-flow (da/en) that an LLM compiles into a clean note, grounded in the live product catalog so it can only name materials that actually exist — a ~60-second flow with deterministic output and a template fallback.
• Material sourcing & supplier comparison. An LLM extracts materials and quantities from plain text, matches them against a multi-supplier catalog, and returns a cheapest-total ranking with delivery estimates — VAT-aware and øre-precise.
• Danish CVR registry lookup. Real-time company verification at signup in a single call from EU infrastructure, normalized and cached for 30 days to stay resilient to upstream limits.
• Over-the-air localization platform. An admin-managed translation store served to mobile via an i18n endpoint with ETag / 304 caching, so non-engineers own the copy and wording ships without an app release.
• Production cloud migration. Moved eight processes to a single Linux host under PM2, Nginx and TLS across three subdomains, with a verified database cutover and no data loss.

The result is a lean, AI-grounded platform running reliably in production:

• 8 services live in production behind one gateway (six domain microservices + two web apps).
• 5 features shipped end-to-end, from AI inference to infrastructure.
• 961 localized strings across two languages, editable over-the-air with no redeploy.