AI Field Support: voice-first, offline-first, Entra SSO, confidence-gated AI

A working voice-first (speech-to-text and text-to-speech), offline-first (service worker plus local-database sync), multi-tenant PWA with real Microsoft 365/Entra SSO plus three other identity providers, an 85%-confidence-gated AI diagnosis path with a live reasoning ledger, and a multi-vendor/multi-model AI failover chain -- backed by two actively developed repositories (4,228 combined commits, oldest dated 2025-08-11, most recent 2026-08-16) with a binding, CI-enforced promotion gate and a documented two-path settings discipline that forces every new setting through a self-heal guarantee. The platform's extensibility is already proven, not aspirational: a second commercial vertical (AI Boat Mechanic) already runs on this same engine via a frontend branding layer, vertical-specific AI prompts, and light API additions, with no fork of the backend, verified at roughly 13 weeks from the marine prompt scaffold to live production, not months or years. Trust-first design carries across the platform too: this engine's marine vertical already gives its diagnostic away on a free tier so an owner can self-assess DIY-vs-shop before calling anyone, and the same free-tier model is planned for this product itself. A technician working on real equipment, often without reliable signal, needs to talk to the tool instead of typing into it, keep working when offline, sign in with the identity system their organization already uses, and trust that an AI diagnosis says "I'm not sure yet" instead of guessing -- addressing a field that widely reports a shortage of experienced technicians and of time to train new ones on the job. A separate, related gap: an equipment owner usually cannot tell a DIY fix from a job that needs a paid shop visit, and often fears being overcharged once they call one. Evidence: speech capture/synthesis, service worker and offline sync, Entra/Azure AD/Google/Apple auth middleware, the confidence-gated analysis service, and the multi-provider AI router, all read directly from source, plus the repositories' own commit history, CRITICAL-PATHS.md (a real-network customer-journey test inventory), the .github/workflows/ gate definitions, and docs/runbooks/settings-add-checklist.md.

Technologies

• Voice input/output (Web Speech API)
• Offline-first PWA with sync
• Microsoft 365/Entra SSO
• Confidence-gated AI diagnosis
• Multi-vendor/model AI failover

Applicability

Useful for a technical or executive evaluator assessing AI Field Support specifically; drawn from internal source evidence, not an independently public-published artifact.

Evidence

Evidence class: delivered-artifact