Physics Ph.D. building simulation-driven insight tools. Semi-retired, running too many experiments in parallel and updating my priors accordingly. I specialize in Bayesian modeling, ML, and simulation as a tool for understanding and decision-making under uncertainty. Lately thinking about how agentic systems change software development, and where the boundary between simulation and inference really is.

import pymc as pm

coords = {
    "interest": [
        "Bayesian Modeling", "Machine Learning", "Simulation", "Computer Science/Type Theory",
        "Gaming", "Music", "Learning New Things"
    ],
}

with pm.Model(coords=coords) as ron:
    # Priors
    hours_in_rabbit_hole = pm.Exponential("hours_in_rabbit_hole", lam=0.1)
    time_allocation = pm.Dirichlet("time_allocation", a=[5, 4, 3, 3, 2, 1, 2], dims="interest")

    # What am I working on today?
    current_obsession = pm.Categorical("current_obsession", p=time_allocation)

    # Likelihood — projects per interest
    fun = pm.Deterministic("fun", hours_in_rabbit_hole * time_allocation, dims="interest")
    projects = pm.Poisson("projects", mu=fun, observed=[15, 8, 6, 4, 1, 3, 8], dims="interest")

    idata = pm.sample()  

📈 rationalpursuit.com — Data analytics consulting: mathematical modeling, Bayesian methods, and ML

🤖 Agentic AI — Claude Code, PydanticAI, OpenRouter — building agents that actually do things

• IslandSim — Agentic AI in a tabletop RPG setting — agents making decisions with limited information (WIP)
• codecrafters-claude-code-python — Built my own mini coding agent from the CodeCrafters challenge

📊 Bayesian & Simulation — My public repos are just the tip of the iceberg. Topics include PyMC, Stan, diffusion models, reinforcement learning, and agent-based modeling.

🎲 RPG & Worldbuilding — Traveller, D&D, Pirate Borg, Dragonbane — player, sometimes GM, always rolling badly

♟️ Chess — admiralentropy on Chess.com — studying openings, tactics, and classic games. One blunder at a time.

🎸 Guitar — Staring to learn .. again. — building a repertoire, one song at a time

📚 Reading — Perpetual student — Bayesian stats, deep learning, agentic AI, and whatever rabbit hole is next

Things I built because I wanted to know how they actually work.

• ParametricEQ — Parametric EQ audio plugin in JUCE (C++)
• EOPLInterp — C++ interpreter from Essentials of Programming Languages
• ToySOCCore — The TOY computer from Sedgewick & Wayne, implemented in VHDL
• nexttoken — Interactive next-token prediction visualizer (Hugging Face). "Vibe coded" using Claude.

"All models are wrong. Some are useful. A few are genuinely fun to build."