Allan ButlerData Scientist

Self-paced course · Python and SQL

Build an analytics agent you can trust

Build a Python agent that can explain a swing in ticket prices, show the math, and catch its own bad query before the answer reaches your team.

See enrollment and pricingTry the free system map

What you’ll build

A local Python agent that translates a weekly search question into a query plan, runs a bounded SQL tool, checks its result, and revises or withholds the answer. You’ll deliver a stakeholder brief alongside the trace that supports it.

Built for data scientists and analysts comfortable with basic Python functions and SQL aggregation. The exercises use fictional retail data, SQLite, and Python’s standard library. The planner runs offline.

Curriculum

Six modules, eighteen lessons, and two applied projects. Open a module to see the assignments.

Module 1Define the question
  1. Turn a stakeholder question into a decision
  2. Define a search that counts
  3. Decide when to ask for clarification

You’ll finish with A metric contract with a denominator, grain, scope, and release rule.

Module 2Build the query tool
  1. Inspect the source rows
  2. Compare three aggregation plans
  3. Package the query result as evidence

You’ll finish with A Python and SQLite tool that returns evidence for two comparable periods.

Module 3Give the agent a repair loop
  1. Give the planner a small action space
  2. Turn failed checks into a repair
  3. Stop when evidence runs out

You’ll finish with A planner that can query, evaluate, revise once, and stop.

Module 4Evaluate the answer
  1. Write an independent reference calculation
  2. Test a plausible wrong answer
  3. Separate a trend from its cause

You’ll finish with A runnable evaluation suite with intentional failure cases.

Module 5Ship the decision brief
  1. Run the holdout scenario
  2. Write a defensible recommendation
  3. Make the release call

You’ll finish with A capstone report, execution trace, and release decision on unseen data.

Module 6Build a sports-market AI workflow
  1. Distinguish the market price from a comparable seat
  2. Route a sports question to the right tool
  3. Evaluate a ticket-market briefing

You’ll finish with A ticket-market analyst with a bounded model planner, a fixed-mix price tool, and a decision trace.

Inside the course kit

An offline course reader with worked examples, assignments and answer checks. You get three simulated datasets, runnable retail and sports workflows, a twelve-test evaluation suite, a capstone brief template, and a grading rubric. The sports workflow includes an optional live model planner using your own API account.

From retail search to the Super Bowl

A ticket desk sees average prices jump while comparable seats get cheaper. Build an AI workflow that chooses the right pricing tool, checks the seat mix, and writes down the evidence. The exercise uses synthetic transactions inspired by my Super Bowl ticket research.

Your instructor

Allan Butler works on applied AI and machine learning in Austin. His writing on search and human review starts with a practical question: what would it take to trust this system with a real decision?

See Allan’s PyData talk →

Course enrollment

$1,500 planned standard $999 launch price

One payment includes the full course reader, both project workflows, datasets, code, evaluations and capstone materials.

Launch pricing runs for the first 14 days of enrollment.

Join the launch list

Get enrollment updates

Leave your email for course enrollment updates from Allan.