CM–AI–01 • case study
BMU Fault Finder.
A source-grounded diagnostic concept for helping façade-access engineers find relevant technical information faster—without handing safety-critical judgement to a machine.
01 / The problem
Field knowledge is valuable, but difficult to retrieve.
Engineers often work across different machines, manuals and fault histories. The useful answer may exist, but finding the right source under time pressure can be harder than it should be.
02 / The concept
Guide the search. Show the evidence. Keep a person responsible.
The prototype structures the reported symptom, retrieves relevant source material and presents possible checks with clear references. It is designed as decision support—not an autonomous diagnosis and never an authority to return equipment to service.
03 / The safety boundary
Useful AI needs explicit limits.
- No invented procedures when evidence is missing.
- Uncertainty and source gaps remain visible.
- Authorised engineers retain all safety-critical decisions.
- Public material contains no client, site or manufacturer data.
04 / What I learned
The interface matters as much as the model.
A strong field tool does not merely generate an answer. It helps the user frame the problem, inspect the supporting evidence, record what was checked and stop when confidence is not justified.