I build AI products that sit close to the real world: voice agents, LLM workflows, evals, and backend systems.
Most of my work lives in the space between a cool AI demo and something that actually works with real users.
I’m building Kalbi.app, a long-distance relationship app that started after I moved from Texas to Virginia and realized long-distance often becomes passive instead of intentional.
I’m also building Burki.dev, a voice AI platform shaped by almost three years of building real-world phone agents. The goal is simple: make production phone agents easier to build, test, and deploy with transparent pricing and no hidden usage mess.
I helped build BiteBuddy.ai, an AI receptionist for restaurants that handled calls, reservations, orders, and customer support.
We grew it to roughly $50k ARR before I moved on to my next projects.
At Texas Tech University, I worked with Professor Akbar Namian on AI alignment research around agent behavior, vague policy following, and deceptive policy behavior in AI tool-calling systems.
The core question was simple: when an AI system is given a policy, tools, and room to act, how do we know it is actually following the policy instead of only appearing to?
I keep coming back to voice agents, evals, observability, tool use, RAG, backend reliability, and AI systems that survive production.
My stack is mostly Python, TypeScript, FastAPI, Postgres, Redis, AWS, Docker, WebSockets, Twilio, Deepgram, OpenAI, Anthropic, and pgvector.