Week 2
Data handling, feature thinking, evaluation, and classical ML before the LLM layer.
This lesson is about data shaping as engineering work, not notebook theater. You are learning how raw tables become trustworthy model inputs.
Checkpoint
ML Decision Boundary GateThis week ends with a gated checkpoint. You progress by shipping a real artifact, not by reading passively.
Deliverable
A simple ML pipeline with evaluation and a leakage audit.Each week leaves behind portfolio evidence that compounds into the final SaaS and its operating narrative.
Week Thesis
This lesson is about data shaping as engineering work, not notebook theater. You are learning how raw tables become trustworthy model inputs.
Bad data silently poisons everything downstream. If your features are inconsistent, mislabeled, or leaky, your model quality and your product decisions become fiction.
Think of the dataset as an interface contract between the real world and the model. Every column carries assumptions about meaning, freshness, allowable values, and transformation history.
This lesson teaches the decision boundary between common supervised learning tasks and the discipline of choosing the simplest model that fits the job.
Lesson Stack
Lesson Preview
Clean data beats clever models.
This lesson is about data shaping as engineering work, not notebook theater. You are learning how raw tables become trustworthy model inputs.
Bad data silently poisons everything downstream. If your features are inconsistent, mislabeled, or leaky, your model quality and your product decisions become fiction.
Think of the dataset as an interface contract between the real world and the model. Every column carries assumptions about meaning, freshness, allowable values, and transformation history.
Lesson Preview
Know when a classical model is the correct tool.
This lesson teaches the decision boundary between common supervised learning tasks and the discipline of choosing the simplest model that fits the job.
Most expensive AI mistakes begin with using the wrong class of tool. Choosing an LLM for structured prediction or choosing a complex model when a baseline would do creates operational waste.
Start with the business question, map it to target type, then choose the least complex model that gives acceptable signal and explainability.
Lesson Preview
A bad evaluation pipeline can make a useless model look great.
This lesson teaches you how to distrust a flattering metric until the evaluation design has earned your trust.
Leaky evaluation produces false confidence, which is one of the fastest ways to launch a bad model with executive approval. Privacy mistakes add legal and reputational cost on top.
Evaluation is a claim about future usefulness. Leakage and privacy failures invalidate that claim by corrupting either the data boundary or the legal boundary.
Portfolio Artifact
audit
A structured audit describing data prep, evaluation design, leakage risks, and privacy boundaries.