Matching Assistant — b*coop
Problem
For every vacant apartment, suitable applicants have to be found and compared across the waiting list and incoming requests — traceably, and without moving more personal data than necessary.
Solution
Authorised staff send the key facts of the apartment — location, rent, rooms, people — via Telegram. The system reads the waiting list and requests strictly read-only from b*coop's existing SharePoint, takes over only the fields it needs and assigns each person a reference; names, contact details and free text stay behind. A hard filter (budget, household size) and fixed scoring rules (location, room need, waiting time, budget) produce the ranking. The AI (Claude Haiku) only phrases a short rationale per match and never sees names — just reference, score and criteria. It never decides, never rejects and never evaluates protected characteristics. A human always makes the decision.
Result
Live at b*coop. Authorised staff get their scored list straight in the chat and look up names and phone numbers themselves in SharePoint via the reference.
Results in Numbers
Always a human
Decision
No names to the AI
Data protection
Fixed, traceable rules
Scoring
Tech Stack
“The citizens' cooperative b*coop commissioned Matthäus Nitz to develop a solution for continuous, automated web research across a range of topics such as funding programmes, job openings and housing offers.
Within a short time Matthäus delivered an innovative and reliable solution that fully meets our requirements. Thanks to the daily updates, we always receive current and relevant information on the topics that matter to us.”

Next Project
Nina — Lobis Böden