FoundationsBeginner

Python for AI Engineering

Six weeks of Python written the way AI systems need it: typed, tested, async where it matters, and structured so an LLM can call it. For people who have programmed before and want the foundation the rest of the curriculum assumes.

College Students Working Professionals

Join the waitlist Get fee details

Duration: 6 weeks
Level: Beginner
Format: Live online
Investment: Request pricing
Next cohort: Dates announced soon

Who this is for

Any prior programming, in any language
Willingness to work in a terminal

This is not for

Anyone who has never written code — start with a general programming course
Experienced Python engineers — go straight to LLM Engineering

What you leave with

A typed, tested Python service that calls an LLM API
A CLI tool with proper error handling and retries
A small library packaged and installed from your own repo

Tools and stack

Python 3.12PydanticpytesthttpxuvRuffDocker

Curriculum

Module 1 — Environments and tooling

Set up reproducible environments with uv
Read a traceback and fix the actual cause
Use Ruff and type hints from day one

Module 2 — Data modelling with Pydantic

Model inputs and outputs as types, not dicts
Validate at boundaries
Fail fast with useful errors

Module 3 — Calling APIs properly

Write a client with timeouts, retries and backoff
Handle rate limits without crashing
Log what actually helps debugging

Module 4 — Async and concurrency

Know when async helps and when it does not
Run concurrent calls safely
Avoid the three common deadlocks

Module 5 — Testing

Write tests that catch regressions, not tests that mirror the code
Mock external services
Use fixtures well

Module 6 — Packaging and shipping

Package a library and install it from Git
Containerise a service
Hand code to someone else without a call

How this program handles evaluation and governance

Responsible engineering starts before any model is involved. This course establishes the habits the later ones depend on: validating at boundaries, failing loudly rather than silently, logging enough to reconstruct what happened, and writing tests that actually catch regressions. Skip these and every governance practice later becomes theatre.