Daniel Rodriguez Gonzalez
DCS · MSDS
Lead Machine Learning Engineer
I build production statistical and machine learning systems, and the verification layer that keeps them trustworthy as they change.
What I work on
Statistical methodology
Survey weighting and raking, calibration, Bayesian inference, and population estimation, formulated as convex optimization where it helps.
Verification for production ML
Preflight checks, distributional drift gates, regression tests for nondeterministic solvers, and reproducible replay. These guardrails make fast iteration, including AI-assisted iteration, safer when correctness is statistical rather than exact.
ML engineering
End-to-end pipelines, orchestration, model versioning, and monitoring on Snowflake, Databricks, and AWS.
How I work
Reproduce before you replace
When I migrate a production system, the new pipeline first has to reproduce the old one’s published output. Only then do I change the method, and I measure what the change does.
Prove the gap, don’t explain it
A reconciliation difference is a bug until shown otherwise. “Solver noise” is a hypothesis to test, not an answer.
Agree on the decision before the code
I put design options and their trade-offs to methodology and engineering partners before implementation, so what ships is what was agreed.
Turn every correction into a rule
I work with AI coding agents under a written playbook. Each rule traces back to the mistake that produced it, and verification is part of every task.
Background
I have spent more than ten years shipping machine learning in media and advertising measurement, defense, and consulting. Today I own production pipelines that calibrate audience-measurement panels to the population. I hold a Doctor of Computer Science in big data analytics; my doctoral research applied residual neural networks to open-set human activity recognition. Before moving into data science, I worked as a field applications engineer for hardware and research organizations.
See Experience for roles, Projects for things I build on my own time, and Research for publications.