Workshop
The Pit Stop
- Where the slow work goes: an ExtendedTask backed by a mirai daemon, off the session's thread
- What the user sees while it runs, and when a stale result is thrown away
Sessions
Recordings you buy once.
Every session is a working practitioner showing the method the way they use it in the job. Buy one and the recording is sent to your account; the Masterclass is every one of them for a year.
Sessions
The 2026 line-up, each session a recording you buy once.
Every one is a working practitioner showing the method the way they use it in the job. Buy a session and the recording is sent to your account.
Workshop
The Pit Stop
Session
KNOW WHEN THE DECLINE STARTS
Know Exactly When an MLB Player Starts to Decline
- How to shape a career over age: a generalized additive model, and where its flexibility is and is not trusted
- How to separate the player from the cohort with mixed effects and partial pooling
Session
CRICKET COMMENTARY INTO DATA
Build an AI Agent That Turns Cricket Commentary Into Data
- The schema first: what a ball, an over and an innings are as rows, before any model sees text
- How to constrain an LLM to extract into that schema with validation on every record
Session
HOT START OR REAL?
Hierarchical Modeling using PyMC in Baseball
- The model: every hitter's rate drawn from a league distribution whose spread is estimated, not assumed
- The priors: what to choose, how to defend them, and how to check them with prior predictive draws
Session
SECONDS, NOT HOURS
Variational Inference in Sports
- When variational inference is the right trade: speed for an approximation you can check
- How to check where the approximation holds and where it understates the spread
Session
The Armchair Scout
- Which features define a player, and how to scale them so no one statistic dominates
- Which distance metric makes sense for the sport, tested against cases you already know
Session
IT RUNS ITSELF
Game-Day Automation
- How to schedule the work: cron on GitHub Actions, secrets handled properly, every dependency pinned
- How the job fails: loudly, with a message that reaches a person, never silently
Session
FILL THE GAPS
The Injury Report Problem
- Naming the missingness mechanism before touching the data, and what it implies
- How to impute with mice, by variable type, and how to check the imputations
Session
KICKOFF TO PRODUCTION
Kickoff to Production in Minutes
- What goes in the brief the agent works from: structure, modules, tests, deploy target
- How to direct the agent and read every file it writes before anything ships
Session
The No-Look Pass
- Returning reactives from a module server versus storing shared state on the session
- When an R6 store is worth it, and when it is too much
Session
ASK THE BOX SCORE
Ask the Box Score Anything
- Where the model sits: writing SQL that DuckDB runs, never touching the rows for a filter
- What the data description must contain for text-to-SQL to be reliable: columns, units, rules
Session
The Farm System
- Drawing the module map first: responsibilities, inputs, outputs
- How to namespace, nest and compose modules so one view can be replaced on its own
Session
What Fills the Seats
- How to set up penalized regression with a split you can defend
- How to tune the penalty by cross-validation and read the whole path, not one number
Session
TRUST THE CHAINS
MCMC in Practice
- What to read in everything pm.sample returns, and which diagnostics decide whether to believe it
- How to recognize the three classic failures, non-identifiability, funnel geometry and slow mixing, and repair each
Session
RUNS ANYWHERE
Dockerizing a Match-Day Shiny App So It Runs Anywhere
- What goes in the Dockerfile for an R Shiny app: pinned packages, a small image, a health check
- How to run it locally with Compose, with data and secrets mounted properly
Workshop
REPLAY THE STAGE
Build a Cycling Analysis Board Without Writing Code
- Which blocks a stage analysis needs: the route and rider data in, the transforms between, the chart out
- How to shape data by snapping transform blocks together instead of writing the pipeline
- Which description of a body the question needs: flat keypoints, a mesh, or a skeleton with joint limits
- What done means, written down before the AI builds anything, and where it must stop
Keep up
One email when something new arrives.
A short welcome note, a few session picks over two weeks, then only the updates you signed up for. Reply stop any time.