Restaurant voice AI / SELECTED ENGINEERING WORK
TREX Voice Ordering Agent
A local-first restaurant voice agent for TFC that answers menu questions and manages pickup or delivery orders.
This overview describes my engineering work on this project, including its purpose, implementation, and technical focus. Project illustrations are abstract; private assets and internal implementation details are not published here.
Overview
A local-first restaurant voice agent for TFC that answers menu questions and manages pickup or delivery orders. My work connects the user experience with the data, application logic, and integrations behind these workflows.
Problem addressed
Voice ordering needs to handle interruptions, inventory checks, and changing customer details while preserving the order state.
Solution
Built streaming speech and structured ordering workflows with Python, FastAPI, WebSockets, Qwen3 8B, and local speech models.
Key features
- Streaming audio and live captions
- Natural endpointing and barge-in
- Menu questions and inventory validation
- Pickup and delivery orders
- Mid-conversation customer updates
- SQLite order persistence
- SOP datasets and Phoenix-compatible evaluations
Technical outcome
A voice ordering workflow with streaming responses, structured order actions, local persistence, and evaluation datasets.
Interested in the engineering behind this work? Get in touch to discuss the approach and a similar product.