Werkstatt Pro: AI parts catalog in a Telegram Mini App

Reduce part lookup for mechanics working in a service bay from 5–10 minutes to a target of 3–5 seconds, without requiring exact article numbers or clean desktop conditions.

Product requirements

• Recognise conversational German workshop language and map it to TecDoc-compatible parameters.
• Support VIN entry, camera OCR and voice input while preserving a practical flow in gloves, low light and unreliable connectivity.
• Keep Telegram authentication and subscription payments inside the messenger flow.
• Meet DSGVO/GDPR requirements for workshop customer data, garage history and audit trails.

Role: product and full-stack architecture — Telegram Mini App UX, AI-assisted search flow and GDPR-aware backend design.

The proposed MVP uses React and TypeScript with @twa-dev/sdk and Telegram theme variables for the client, plus an async FastAPI service, Celery and PostgreSQL. GPT-4o mini normalises spoken German requests into structured TecDoc API searches; Whisper handles speech-to-text and the camera flow supports document OCR. A local “virtual garage” caches five recent VINs, while Stripe Checkout or Telegram Stars covers subscriptions. The interface prioritises large touch targets, monospace part numbers, one-tap copy and explicit loading states.

MVP targets

• Time-to-part ≤ 5 seconds; TMA load time ≤ 1.5 seconds.
• AI recognition accuracy ≥ 92%; VIN validation success ≥ 98%.
• Trial-to-paid conversion ≥ 15%; 7-day retention ≥ 60%.
• GDPR compliance 100%; offline cache-hit rate ≥ 30%.