Michał, the after-hours waste hotline

Field notes

The knowledge base reportedly came from the waste company’s own service rather than being generated solely by a language model. Its checking process covered 70 pages, extracted 372 amounts, phone numbers, hours, addresses, and limits, and found 13 errors before deployment. The demo also describes a complaint workflow that confirms the resident’s address and waste type before requesting a callback number and adding the case to a submissions list.