Week 5 (AI track): Guided build#

← Back to Weeks

Goal#

A small app around a model, plus an intro to retrieval, with instructions that are still fairly step-by-step.

Scope#

• A summarizer, classifier, or chatbot with real memory, built around a hosted model's API.
• Retrieval, conceptually: why raw model calls aren't enough for anything that needs facts, what a vector search does.

The project: Notes Assistant#

This week (and Weeks 6-7 after it) builds one continuous project: Notes Assistant, a command-line tool that summarizes personal text notes and answers questions about them using a simple, from-scratch version of retrieval. It's chosen specifically because it maps onto the skills the Week 8 "docs assistant" capstone needs (retrieval, source citation, refusing to guess when the documents don't cover a question), without building that capstone itself.

Do this in order#

Steps 1-4 are required. Everything after is optional extra practice; skipping straight to Week 6 without it leaves no gap in what Week 6 assumes you know.

1. notes/01-why-retrieval-matters.md through 03-organizing-notes-assistant.md, in order.
2. exercises/00-guided-build-retrieval.md, a line-by-line guided build of the retrieval core (no API key needed), before the project walkthrough.
3. exercises/01_single_prompt.py, 02_multi_turn_chat.py, 03_structured_extraction.py, in order: the API-shape progression from a single call to real memory to structured output.
4. project/README.md's full walkthrough of the finished notes_assistant.py.
5. Optional: exercises/04-guided-build-citations.md, a second guided, line-by-line build (also no API key needed), the citation-formatting logic answer_question depends on.
6. Optional, any order: exercises/05 through 11, further standalone reps (naive substring matching vs. keyword overlap, a second chunking strategy, adding your own sample note, chunk overlap, stopword tuning, top_k tuning, whether the system prompt is really grounding the answer). Most need no API key; see exercises/README.md for exactly which ones do.
7. Optional: project/bonus_vector_similarity.py, a real vector and similarity formula in plain Python.

Status#