No-Show Desk: TabPFN spots the motel guests likely to cancel and Gemma asks them first

Описание и применение

Небольшой мотель использует систему No‑Show Desk, где модель TabPFN оценивает вероятность отмены бронирования, а Gemma генерирует сообщение‑подтверждение. За 30 дней система выявила 111 из 172 отмен, в то время как случайный выбор выявил около 64.

Задача

Small motels lose revenue when guests do not show up and no deposit was taken.

Результат

Over 30 test mornings, the system caught 111 of 172 cancellations (vs ~64 for random picks). Drafts with incorrect facts were 13/100 with Gemma, 0/100 when checked by code. Message generation time was 3.25 s vs 31.4 s for a local Ollama build.

Рабочий процесс

• Receive Telegram message at 08:00
• Open owner page and view list of risky guests
• Click a guest row to see booking details and risk explanation
• Press "Draft a message" to generate a confirmation email
• Review draft and confirm facts are correct
• Send email or skip sending

Доказательства

Scored on 30 real past mornings, learning only from earlier bookings, its picks caught 111 of 172 cancellations, against about 64 for random picks of the same size.

Источник

Drafts with a wrong date, night count or total 13 of 100 when Gemma worked them out, 0 of 100 when code handed them over

Источник

Seconds per message, Gemma in the browser against Ollama (a program that runs models locally) on the same Mac 3.25 against 31.4

Источник