adrian tame

I am an AI/ML engineer who builds tools for and with LLM agents. I am currently building Mollify, a Rust-native static analysis engine that gives coding agents deterministic ground truth about Python codebases, and Foothills Labs, an AI foundation lab whose first projects are labloop, an agent-driven experiment loop, and regexbench, an evaluator for LLM-generated regular expressions. At work I focus on applied ML, including structured document extraction with OCR at scale.

Underneath the agent work is a decade of statistical practice: Bayesian optimization (my MSc thesis develops Bayesian Adaptive Spline Surfaces as a surrogate model and an alternative to Gaussian processes), time series forecasting in finance, causal inference, and large-scale experimentation and A/B testing. The through-line in both halves is the same: I care about systems whose conclusions can be trusted, with honest uncertainty, tiered evidence, and results that survive contact with production.

Right now I am finishing the thesis, and along the way I am learning agentic harness design, econometric theory, and Rust. I am always happy to collaborate on agentic systems, time series forecasting in finance, and anomaly detection.