Antonis Mavritsakis

I help technical teams scope and build ML and AI systems. My focus is on problems where the modeling is genuinely hard: reinforcement learning, probabilistic and Bayesian models, retrieval-augmented generation over technical corpora, custom optimization, and game-theoretic decision-making under uncertainty.

What I bring to a project is honest scoping and the ability to tell you when ML is the wrong tool. Most engagements start with a working session where we map your problem to the right approach, identify the highest-risk assumptions, and set realistic timelines. From there I either implement the system end-to-end, audit and improve an existing pipeline, or hand off a scoped technical spec your in-house team can execute.

What sets me apart is depth in niches most ML generalists avoid. I have built benchmarked CFR-based poker solvers, designed reinforcement learning environments, and shipped RAG pipelines for specialized technical domains. I am as comfortable in research-grade math and developing novel solutions as I am in production Python.

Typical deliverables include scoping documents and technical specs, end-to-end ML pipelines, reinforcement learning training systems, custom solvers and optimization code, RAG and LLM-driven applications, and code or architecture audits with prioritized recommendations. I work best with clients who value precision over hype and want a collaborator who will push back when the plan does not make sense.