Full Stack Engineer
Full stack engineer building the layer where people and AI meet.
Building multi-agent systems, the harnesses that make their output reliable, and streaming AI interfaces, on Next.js, React, and TypeScript.
Open to remote roles · US, Eastern Time · No sponsorship needed
About
What I do
I'm a full-stack and agentic engineer. For the last four years at Caqophony I've shipped four production products end to end, from the React interfaces people touch down through the APIs, the data, and the model layer underneath.
My deepest work is in agentic systems: multi-agent pipelines that turn a plain intent into working software, the harnesses and evaluation loops that keep their output reliable, and the cost-aware model routing that makes them affordable to run. The throughline of that work is a conviction that quality lives in the system around the model, not the model itself. A good harness gets near-perfect output from whatever model is current, by constraining what the model is free to get wrong and treating every derivable decision as engineering rather than guesswork.
The range is real too. I've built streaming AI interfaces, peer-to-peer marketplaces, commerce and payments, mapping and search, all on Next.js, React, and TypeScript with production discipline throughout. I move fast, I care about code that holds up under its own tests, and I stay close to where the tooling is heading, since most of what I reach for now is newer than a year old.
Skills
What I work with
My deepest area. I build multi-agent pipelines, the harnesses and evaluation loops that keep their output reliable, prompt and context engineering, tool use over MCP, and sandboxed execution with cost-aware routing. The point is to build the system around the model, so quality never depends on the one you run.
- Multi-agent
- Harnesses
- Prompt engineering
- Evals
- Claude / OpenAI
- MCP
- E2B sandboxes
- Cost routing
My deepest area. I build multi-agent pipelines, the harnesses and evaluation loops that keep their output reliable, prompt and context engineering, tool use over MCP, and sandboxed execution with cost-aware routing. The point is to build the system around the model, so quality never depends on the one you run.
- Multi-agent
- Harnesses
- Prompt engineering
- Evals
- Claude / OpenAI
- MCP
- E2B sandboxes
- Cost routing
Work
Selected projects
Contact
Get in touch
The fastest way to reach me is email. I'm open to remote roles, US based on Eastern time, and available from early August.