All work04
Teams-BOT
A German-language equipment assistant inside Microsoft Teams that combines inventory lookup, document retrieval and advisory planning in one service.
- Year
- 2025 — 2026
- Role
- AI Developer
- Client
- RÜKO GmbH Baumaschinen
- Stack
- FastAPI · LangGraph · Gemini · PostgreSQL · Pinecone · Redis
01 — Problem
Equipment data for thousands of construction machines was split between structured inventory, manuals and specialist knowledge. Answering a field question meant switching tools, knowing the database, or finding the right colleague.
Employees already worked in Microsoft Teams, so the answer needed to arrive in the conversation — with the system choosing the right knowledge path behind the scenes.
02 — Approach
I built a FastAPI service around the Teams webhook and used LangGraph to route each request to the right capability: PostgreSQL inventory lookup, Pinecone retrieval over manuals, or Gemini-based advisory planning.
The orchestration keeps structured facts and document evidence separate until answer composition. Redis preserves per-thread context when available, while health, reset and admin surfaces make the service operable beyond the happy path.
The result is one deployable backend for direct equipment queries and open-ended project recommendations, rather than a collection of disconnected demos.
03 — Result
Equipment knowledge covering 2,395+ indexed records is available from the Teams conversation employees already use. The same assistant can move from an exact machine lookup to supporting documents or a broader advisory request without changing tools.
