book-setup

name

book-setup

description

Orients readers in the companion repository of the book "Applied AI in Cyber Threat Intelligence" (Packt): chapter folder layout, the per-chapter venv and requirements pattern, GOOGLE_API_KEY .env setup and costs, which chapters need no key, recovering shipped canonical results with git checkout, and how repo code relates to the printed book. Use when getting started with the repo, choosing a chapter to work on, or for setup questions not specific to a single chapter lab.

Book Companion — Repo Orientation

Each chapter-NN/ folder is a self-contained lab with its own README, requirements file(s), and code. Every chapter with a lab has a dedicated skill — read that chapter's skill before doing chapter-specific work; this skill covers only what applies repo-wide.

Chapter	Lab shape	API key?	Dedicated skill
4	Python-foundations notebook	No	ch04-python-foundations
5	Concept figures only — nothing to run	—	—
6	ADK agent + grid-search notebook	Yes	ch06-requirement-evaluator
7	Collection pipeline + eval notebook	Yes	ch07-collection-lab
8	Processing pipeline + eval notebook	Yes	ch08-processing-lab
9	Analysis (SAT) pipeline + eval notebook	Yes	ch09-analysis-lab
10	Production pipeline + eval notebook	Yes	ch10-production-lab
11	Dissemination pipeline + eval notebook	Yes	ch11-dissemination-lab
12	Feedback pipeline + eval notebook	Yes	ch12-feedback-lab
13	Classical-ML notebook + pinned datasets	No	ch13-cti-ml-lab
14–16	Standalone tools shipped as zips — unzip, follow inner README	varies	—

Universal facts

• Environment: fresh venv per chapter inside its folder; pip install -r requirements.txt (+ requirements-simulation.txt where present). Notebooks launch with jupyter lab "<notebook name>".
• API key (ch 6–12): .env per the chapter README with GOOGLE_API_KEY='...' and GOOGLE_GENAI_USE_VERTEXAI=0; free tier at https://ai.google.dev/gemini-api/docs/api-key. In chapters 6–8 (plus chapter 10's restored grid and demo cells) the paid notebook cells carry a # NOTE: REQUIRES API KEY comment with call counts; other cells in chapters 9, 11, and 12 are not individually labeled — use each README's "Cost expectations" note (every keyed chapter has one) before running evals.
• Canonical results are git-protected: chapters ship recorded results (pre-filled CSVs, checkpoints, figures) that some eval cells overwrite. Recover anything with git checkout -- <path>; copy files aside before running evals when the reader wants to keep the canon.
• Recorded results are allowed to fail: chapters 8 and 10 ship runs that miss their own thresholds by design (see their READMEs' "A Note on the Recorded Results") — a reproduced FAIL is not a setup error.
• The repo is newer than the printed book: post-print fixes live here; where a printed listing differs slightly from repo code, the repo is the corrected version. Bugs deliberately kept to match print are listed in each chapter skill under "Known issues frozen in print".
• Going beyond the labs: every chapter skill has a "Directions for improvement" section grounding the chapter's own stated next steps (re-run grids under current models, close recorded eval gaps, recalibrate judges, extend catalogs/ground truth, adapt to enterprise data) with the eval harness to measure against and the canon-protection steps to take first.