I'm Cesar Garza, a senior AI systems and infrastructure engineer in Texas. I build and ship production ML systems end to end — the models, the LLM and agent infrastructure around them, and the deployment and observability that keep them honest. I design for the operator: the safe path should be the easy path.
Some of what that looks like in practice. I built Mandate, a governed agent control plane that runs LLM agents as leased, budgeted, audited jobs — policy enforcement, approvals, and output guardrails live in the platform rather than in the model's tool calls — deployed on Kubernetes and continuously verified in production. I built SplatGPT, an 83M-parameter set-completion transformer with sparse-autoencoder interpretability, written up on this site as a case study series. I designed a graph-based rating engine over 75K+ event outcomes, validated against independent peer voting. And I'm a co-inventor on a published US patent application for multi-agent LLM coordination. By day I do data-science engineering at Enverus.
This site first went up in June 2018 and has worn a few names since; these days it just carries mine. I write essays and open-notebook posts about the systems I build — the reasoning and the dead ends, not just the results.
Outside of work I write poetry. It lives at poetry.cegarza.com.