Senior Software Engineer · Tri-Cities, Tennessee

I build systems, then preserve what they teach us.

I'm William “Billy” Bratz. For 11 years, I've built distributed systems across identity, underwriting, analytics, and conversational AI. Today my focus is production LLM agents, MCP infrastructure, and the context those systems need to make sound decisions.

William Bratz

What I work on now

Making capable AI systems dependable in production.

The interesting part of AI engineering begins after the demo works. Production introduces real users, imperfect data, latency budgets, failure modes, shifting models, and consequences. I work on the tool, context, evaluation, observability, and reliability layers that help agent systems operate inside those constraints.

I'm especially interested in the boundary between product intent and engineering implementation. When the reason behind a decision disappears, both people and agents are forced to reconstruct it. Better context makes that reasoning durable, reviewable, and useful the next time a decision reaches code.

How I work

A few principles that keep showing up.

01

Evidence beats confidence

Good decisions should survive contact with logs, traces, source material, and the people affected by them.

02

Production reveals the system

Architecture diagrams describe intent. Production behavior shows where the real boundaries, dependencies, and risks live.

03

Context should compound

Decisions, incidents, and implementations should make the next piece of work easier, not disappear into another chat or ticket.

Work made public

Ideas become more useful when other people can test them.

Independent products

Small tools can still make an idea tangible.

Free to use

Useful experiments, released into the world.

Neverending Story Pointer

A free planning-poker app for teams that want a quick, shared way to estimate work without turning the ceremony into the work.

Open the pointing app

How Unique Is a GUID?

An interactive explanation of an enormous probability space, with collision calculators, relatable comparisons, and a secure v4 generator. The same experience is available in GUID and UUID vocabulary.

Let's compare notes

Working on a hard system or a better way to preserve what your team knows?

I am always interested in thoughtful conversations about production AI, MCP, distributed systems, and durable engineering context.