Software engineer, San Francisco
I build AI that real organizations can trust. Most recently I was a Solutions Architect at RespondIt AI, deploying voice agents for public housing authorities and property operators, from the first discovery call to the report their leadership reads. On my own, I build Aura, which covers about 30,000 U.S. places and writes an environmental brief for any of them on demand, each locked to the data behind it. Before that, I built backend and cloud systems at AT&T.
housing and property organizations I've deployed or supported
Where this comes fromOrganizations where I built the agent, supported a live deployment, or ran a pre-deployment trial, May 2025 to September 2026. Sales demos alone aren't counted.
calls a month handled by the agents at the flagship client
Where this comes fromInbound calls on the flagship client's main line, from call exports I reconciled against the product dashboard, April to August 2026.
AI-created work orders I found silently failing, escalated as P0
Where this comes fromWork orders at a large housing authority that the agent confirmed but that never reached the property system, over a 30-day window in 2026. Approximate.
Each of these numbers has a source. Tap one to see it.
An AI agent told callers their repair was booked. About 1 in 10 never reached the property system. I traced why and designed the fix.
RespondIt AI, 2026
Read the case studyLLM-drafted client reports started inventing figures and causes. I built a reporting system where every number has to trace back to the data.
RespondIt AI, 2025 to 2026
Read the case studyIndependent project, 2026
Covers about 30,000 U.S. places and writes an environmental brief for any of them on demand. Aura Local is the city view; Aura Global tracks planetary indicators. The model's prose is locked to data badges computed in code, then audited before anything publishes.
React · Node.js · PostgreSQL · Claude API · 20+ public data sources
Independent project, 2026
Sun or shade, jacket or shorts. A microclimate forecast for 15 San Francisco neighborhoods, with separate sun and shade readings from apparent temperature.
React · TypeScript · Node.js · Open-Meteo · Census Geocoder · Steadman apparent temperature
May 2025 to October 2026, San Francisco
I sat between the client and engineering. I learned how each housing authority actually ran: its phone lines, its property systems, its policies, the questions residents called about. Then I turned that into an AI agent that worked there, and stayed on to measure whether it did.
Built an AI-assisted vulnerability remediation tool, MuleSoft integrations between legacy and modern systems, and Terraform pipelines on Azure for 5G network functions.
Software Engineer and QA Analyst, part-time
Moved messaging services from Python to Go and built GCP and Firebase backends for Flutter apps.
B.A. in Sustainability Studies, with the Elements of Computing certificate