I build production AI systems.
I also build the teams that ship them.

Twenty years in technology. Fifteen shipping mission-critical systems. Current focus: large language models, agentic AI, and the engineering enablement that lets a team move twice as fast next quarter.

About

I am a Principal Software Engineer based in California. Most recently I led engineering on a 0-to-1 multi-tenant AI communication insights platform. Now looking for the next interesting project and a long-term team. I want to solve hard problems people care about.

What I have shipped

Production AI built on Graph RAG, LLMs, and agentic workflows
Event-driven, serverless systems on AWS and Azure
A React-based instant claim payout system that moved well over $1M
Confidential compute on Intel SGX for medical AI use cases
Multi-sensor data capture for Reality Labs at Meta

How I work with teams

Six years as a people manager. Five-plus as a technical lead
Grew an architecture practice from four to eleven engineers in two years
Hundreds of engineers and aspiring engineers mentored over the years. A few examples:
- Mentoring a security professional through the move to independent consulting
- Helped a senior engineer retool for the AI era
- Coached a new engineer into becoming a celebrated lead
- Introduced a teenager to game development. They went on to major in computer science
Repeat speaker at high school career days

Where I show up publicly

LinkedIn: I share current events in AI, security, and software architecture, with my own writing mixed in (linkedin.com/in/charlescozad)
Substack: long-form essays for readers who want the full argument, not just the takeaway (charlescozad.substack.com)
Podcast: guest on the Build Different Podcast, episode 7 “What is Simplicity”, on why simple solutions are the hardest to build and how young engineers should approach AI tools
Open-source reference designs at github.com/ccozad/ml-reference-designs

What I am working on

Three problem shapes I keep coming back to:

Production AI systems. Graph RAG. Agentic workflows. Evaluation frameworks that grade output by code, by model, and by human, each catching a different class of mistake.
Async offload for long-running operations. The pattern that accepts a user request in under a hundred milliseconds and runs a five-minute job behind it without losing the thread. This one comes up in conversations more than any other backend topic right now.
Engineering enablement. Pairing on architecture. Code review with intent. Test-driven development. Clean statement-of-work hygiene. The unglamorous habits that compound across a team.

Seven demos that show this in motion. Explore each in your browser.