Phase 1 — Foundations

Week 1 — Model Mechanics

Map tokens, context windows, attention, decoding, and inference constraints. Build a tiny token and latency notebook. Explain why fluent output is not verified knowledge.

Artifact: model behavior lab with five controlled experiments.

Week 2 — Prompts as Interfaces

Separate policy, task, context, examples, and output contract. Practice instruction hierarchy, delimiters, and structured responses.

Artifact: versioned prompt module with schema validation and failure cases.

Week 3 — Embeddings & Search

Compare sparse and dense retrieval. Study distance metrics, chunking, metadata, ranking, and recall.

Artifact: small search benchmark with labeled queries and top-k analysis.

Week 4 — Evaluation Basics

Turn desired behavior into cases and rubrics. Combine deterministic checks with human review. Learn to inspect distributions and slices rather than only averages.

Artifact: a 25-case task suite with baseline results and an error taxonomy.

Phase checkpoint

You can describe the major sources of uncertainty, design a bounded model interface, retrieve relevant evidence, and measure a change against a stable baseline.