~ $ cat demos/spatial/README
spatial knowledge layer
A chatbot that knows one university campus room by room, and has to show its working for every answer. I'm still finding out whether it works at all.
// what is this
The campus is data: buildings, floors and rooms from OpenStreetMap, plus the equipment in them and its manuals. Ask where a room is and how to get there without stairs, or what's installed somewhere and what the manual says about it.
Who you pick matters. A visitor only sees the public parts, staff see the rest of the rooms, and maintenance also sees the plant and kit behind locked doors. Same question, different answer, and nothing you type can talk it into seeing more.
Most of the equipment is made up. OpenStreetMap has the rooms but not what's bolted to their walls, so I generated some to have something to ask about, and the answer tags it generated when it leans on that. Every sentence also has to point at a record or a passage, and a second check throws out the ones that don't hold up. So sometimes it tells you it can't answer instead of guessing. That's on purpose.
Three questions per conversation, then start a new one.
snapshot 2026-10-24
pick who is asking: