Built for a client
Valley O'Ville AI Operations Agent
An AI agent the resort is run from, in Telegram: bookings, payments, rates, guest replies, hiring and Facebook posts, with nothing saved until a person says yes. It now also answers guests' questions through a chatbot on the resort's website, and reads the resort's Gmail to run hiring end-to-end.
- Valley O'Ville Family Resort — the people running it day to day
- AI Systems Engineer
- February 2026 – Present
The situation
Running the resort meant keeping one official booking record right across a spreadsheet, a calendar and guest folders, while also quoting rates, chasing payments, answering guests and keeping the Facebook Page alive. Every one of those was a separate place to check by hand, and a missed step meant a double booking, a wrong balance or a guest left waiting.
What I built
It started in February 2026 as a simple Telegram bot that filed bookings. In September I rebuilt it as a full operations agent. Staff ask it things the way they would ask a colleague, and it answers from the live booking record, not from memory. When something needs to change — a new booking, a payment, a date move — it first shows exactly what will change, money before and after included, and only saves after someone replies yes. The same booking rules also sit behind a simple Book button app inside Telegram, so a form and a conversation can never disagree. Guest messages are drafts for a person to send, and Facebook posts go out only after an explicit confirm. In October 2026 the agent's reach widened to the resort's Gmail: it now finds and labels every job application, reads each resume in whatever format it arrives, cross-verifies every extracted fact against the resume itself, and maintains a scored hiring tracker in Sheets with the resumes filed in Drive. Later that month it moved to an always-on cloud server, so it keeps working when my laptop is closed, and it backs itself up every night.
Built and maintained alongside my role as the resort's General Manager & AI Systems Engineer, so it was designed against the real workflow, not a guess at one. Some newer pieces — checkout reminders, photo uploads from a phone, and turning handwritten guestbook pages into testimonials — are still being proven in daily use.
What it lets people do
- 01Check a booking, today's arrivals and departures, or whether dates are free, just by asking
- 02Record bookings, payments, charges, deposits, cancellations and date changes, each previewed before it is saved
- 03Get a price quote from the resort's seasonal rates and add-ons, and apply loyalty rewards
- 04Draft guest replies, with separate rules for direct and Airbnb guests
- 05Answer guests' questions directly through the agent's extension to the resort's website, a chatbot that is now live
- 06Run hiring from the inbox: find every application, read every resume format, keep a verified scored tracker in Sheets and the resumes filed in Drive
- 07Prepare Facebook posts and read Page results, publishing only after an explicit confirm
- 08See who changed what, when and why, in a full change log
- 09Reach the same agent any time from Telegram, the desktop app or a terminal; it runs on an always-on cloud server
Featured in
Marc is also a contributor to NYOTesting and feedback, internal code, and UI/UX work on the LLM router this runs on. (opens in a new tab)
Under the hood
The tools this is built with. You do not need to know any of them to use it — open one if you are curious what it does.
- Oracle Cloud (Ubuntu, ARM)