Housekeeping GM co-pilot
The housekeeping manager texts a bot in plain English to build and rebalance the day's room assignments, and it answers back.
Building the daily housekeeping board, meaning which housekeeper cleans which rooms, was a manual and uneven job. One person could end up with 270 minutes of work while another had 100. There is no housekeeping-ops manager sitting above every property to even it out, so it drifted.
How it works
It is a Telegram bot running always-on on the mini. It pulls the day's rooms and the crew actually clocked in, then runs a credit-balancing engine that assigns rooms so everyone's workload lands inside a tight band. In testing it cut the spread from 170 to 270 minutes down to 15 to 25.
- Deterministic path: slash commands and plain-English edits like "move 214 to someone else," handled by an NLU layer, fast and predictable
- Free-text path: anything else bridges to headless Claude for a fluent answer
- Crew comes from live clock-in data, rooms from the housekeeping board
Stack: a Python assignment engine, the Telegram Bot API, the Claude Code CLI on the mini for the conversational fallback, scoped to a single trusted chat, with launchd keeping the poller alive.
It is deployed and running. It is also the front end for a bigger bet: eventually retiring the paid housekeeping app and running that layer myself. A manager with a phone and a bot is the workforce-ops seat that used to be a person with a clipboard.