Building AI Voice Assist: bridging Twilio voice streams with OpenAI's Realtime API over WebSockets — sub-second audio, barge-in handling, in production.
A decade of telephony concurrency translates directly to real-time voice failure modes. The institutional knowledge lets me debug a seven-year-old race condition the same afternoon I ship a new feature on top of it.
Operating as architect / reviewer / integrator — directing AI coding agents on implementation, owning the parts that have to be right.
session_2● DISCONNECTED
Oct 2022 → Dec 2024
Staff Engineer@Grayscale Labs[Atlanta, GA]
stack: [Ruby on Rails, PostgreSQL, React, SMS pipelines]
High-throughput SMS automation for recruiting — the kind of system where one bad query stalls a queue feeding live candidate conversations.
Owned the messaging pipeline through several traffic regime changes.
session_3● DISCONNECTED
Oct 2014 → Aug 2022
Staff Engineer@CallRail[Atlanta, GA]
stack: [Ruby on Rails, PostgreSQL, Redis, Sidekiq, AngularJS]
Owned the code routing hundreds of thousands of daily phone calls, including the race conditions and concurrency bugs that come with it.
Rewrote the 2,000-line ERB + JavaScript tracking library distributed to customers, converting it to pure JavaScript with a comprehensive test suite.
Leveraged the new test suite to remove jQuery from the tracking library, dramatically reducing bundle size.
Progressed from full-stack AngularJS + Rails work to backend-focused systems development.
AI that calls your family and shares summaries. Bun, Elysia, Twilio, OpenAI. Same Twilio ↔ OpenAI Realtime pattern as Voice Assist, run solo end-to-end.