Agentic Data Science
What this course is
The staff a research lab runs on — the data manager who guards the raw files, the methodologist who writes the SOPs, the RA pool that does the legwork, the referee who checks the result — are people you cannot hire alone. So you configure them instead. You teach coding agents your project, bind them with rules you write, and audit their work to the last row count. This is a course in staffing a lab out of agents: turning Claude Code and Codex into the colleagues a one-person shop never had, bound by guardrails you can read and reproduce.
The case study
One question runs the length of the course: when the sky turns, who still rides? Every lesson works the same evidence — the NYC yellow-taxi trip records (TLC, 2024) joined to hourly weather (Open-Meteo) — building toward a model of how a storm reshapes a city's demand. It is a real, awkward, multi-source dataset: the kind a team of analysts used to spend a quarter wrangling.
Get the data → — one command pulls the fixed slice, no API keys.
The route
Six landmarks, sixteen stops. Each capability arrives the moment the growing lab needs it.
- Unit A The Harbor — Day One Arrival; you hire your first hand at the dock.
- Unit B The Library — Onboarding The lab manual written; a reproducible home; the rare-books vault fenced.
- Unit C The Waterworks — The Methods SOPs codified; contracts filter at the pipe; plumbed into the data systems.
- Unit D Grand Central — The Lab at Scale RAs dispatched; parallel tracks; overnight trains; the fleet timetable.
- Unit E The Lighthouse — Autopilot The beacon mechanism; the keeper’s schedule; the lamp design, shipped.
- Unit F The Printing House — The Payoff The press that runs itself — research orchestrated on a loop, refereed, shipped as a report.
How it plays
- A map you ink as you travel — each stop you finish is stamped onto the parchment.
- Terminal simulations of real agent sessions — watch Claude Code and Codex work, then take the keys.
- Quizzes that stamp a wax seal when you pass — the milestone marks on your route.
- A field journal that keeps your running record: incidents, prompts, notes, hours.
It is a course wearing the clothes of an expedition game.
Time & outcomes
Sixteen lessons, about 14 hours end to end — the beginner ramps at the Harbor rise to a 2½-hour capstone at the Printing House.
By the last stop you will have built, for your own project:
- a lab manual the agents actually follow;
- methods that enforce themselves;
- a parallel fleet of agents working at once;
- an unattended pipeline that runs overnight;
- a report that ships itself.