Access-controlled prototype
AI Voice Booking Assistant - Emma
A real-time voice booking prototype combining conversational AI, validated booking tools, persistent storage and a staff dashboard.
01 / Problem and approach
The problem
Explore collecting restaurant reservation details through natural spoken conversation while keeping booking rules deterministic.
My contribution
Connected browser audio, speech recognition, an LLM, speech generation and booking tools. Built contact validation, SQLite persistence, a staff dashboard, access controls, rate limits, call-duration controls and logging for reply delay, token usage and estimated call cost. Deployed with Docker and Traefik.
How it works
- Speak to Emma in the browser in English.
- Booking tools validate the requested details against deterministic rules.
- Confirmed bookings are stored in SQLite for staff review.
- Requests outside the implemented scope are handed to staff through callbacks.
02 / Evaluation and evidence
- The repository documents offline automated tests and eight live AI conversation scenarios. No pass rate or measured latency is claimed here.
03 / Scope and context
Scope and limitations
Private source · Demo available on request. Phase 1 supports English browser conversations and local booking storage. Changes, cancellations and unsupported languages go to staff callbacks. Telephone connectivity, SevenRooms integration, direct changes/cancellations and multilingual voice booking are not implemented in the reviewed version.
Project chronology
A design specification dated 27 October 2026 establishes documented work by that date, not the exact project start.