THE PRODUCT · FOUR ROTATING TRACKS

Four tracks. One is aimed at exactly where you're stuck.

Each cohort runs one track, live and capped. Pick the one that names your problem, the goal is always the same: get you shipping.

01 · ML Basics 02 · ML Engineering 03 · Computer Vision Data Engineering 04 · Agentic Engineering

01

ML Basics

FOR: CONCEPT → FIRST REAL MODEL

For people who understand ML conceptually but have never trained a model or shipped anything that runs. We close the gap between reading about it and having built it — including how today's models like LLMs and VLMs actually work.

Train your first real model

End to end, not a toy example.

Read the signals

Know when a model is actually working.

Ship something that runs

Leave with a working artifact.

Build real intuition

Code-first, from scratch.

02

ML Engineering

FOR: NOTEBOOK → PRODUCTION

For people whose models live in notebooks and never make it to production. We take the thing that works on your machine and make it hold up in the real world.

Package & serve

From script to a callable service.

Handle the failure modes

The ones nobody warns you about.

Monitor in production

Know when it drifts or breaks.

Own the whole path

Data in, prediction out, reliably.

03

Computer Vision Data Engineering

FOR: MESSY IMAGE DATA → CLEAN SETS

For people losing time to messy, duplicate-ridden image data before any vision model can even be trained. The unglamorous work that decides whether the model ever gets good.

De-duplicate at scale

Find near-duplicates fast.

Build clean pipelines

Repeatable, not one-off scripts.

Label & audit quality

Trust the set before you train.

Version your data

Reproducible vision datasets.

04

Agentic Engineering

FOR: DEMO → ARCHITECTURE THAT SHIPS

For engineers stuck at fragile AI-agent demos who need architecture that ships, including agentic coding workflows. The one track also open to non-technical people who want to build with agents.

Design for reliability

Agents that hold up past the demo.

Tools, memory, control

The real architecture, not prompts.

Evaluate & guardrail

Catch failures before users do.

Open to non-technical

Anyone who wants to build with agents.

THE MODEL

How a cohort actually runs

100% live

Every session is real-time. Show up or miss it.

One / 2 months

One cohort every two months, one track at a time.

Capped seats

Strictly limited. Scarcity is by design.

2-week window

Enrollment opens only in the final two weeks.

Find your track live before you enroll.

The free weekly workshop is the fastest way to see which track fits, and how we teach.