casebooks

Searchable library of 20 MBA consulting casebooks, plus a mental math drill.

I built this for my girlfriend while she recruits for MBB consulting, and she uses it regularly. Finding a case by industry or concept takes seconds instead of flipping through PDFs, and Case Math Drill covers the mental-math half of prep.

→ casedrills.vercel.app

how it works

casesearch greps the full text of every book and prints the school and exact PDF page; a second argument restricts it to one school. caseask is for when you can't remember the wording: it searches by meaning, entirely on your machine, no API keys. render-pages Columbia_2021 144 146 turns exhibit pages into PNGs and opens them.

Case Math Drill is the practice side: timed questions across 20 topics, from multiplication and percentages up to break-even, CAGR, NPV and LTV, each with a time limit and a method hint. Per-topic accuracy is saved in the browser.

The books come from Hacking the Case Interview. What each one contributes is in COVERAGE.md, and the math from all 20 is distilled into case_interview_math.md.

run it locally

git clone https://github.com/joudbitar/casebooks.git
cd casebooks
./download_casebooks.sh      # fetches the 20 PDFs
./scripts/extract_text.sh    # builds the text index (needs poppler's pdftotext)

The first ./caseask run creates a Python env with uv and builds the embeddings index. That takes a few minutes, once.

how it's built

Two bash scripts, two Python scripts, and one HTML file. No server, no database.

• casesearch is a single awk pass. pdftotext separates pages with form-feed characters, so counting them while matching maps every hit back to its real PDF page. The "index" is 20 plain text files.
• caseask embeds 4,013 page-sized chunks with model2vec's potion-base-8M, a static embedding model that runs locally with no GPU, and ranks queries by cosine similarity.
• Case Math Drill is one HTML file with zero external references. vercel.json is four lines: serve the app directory, no framework.
• The PDFs are not committed. download_casebooks.sh rebuilds the whole library from source URLs.

what it doesn't do

• Built on and for macOS. render-pages opens Preview, and the scripts assume poppler and uv are installed. Linux is probably close but untested.
• casesearch is a grep, not a search engine. No ranking, no fuzzy matching.
• Drill progress lives in localStorage. No accounts, no sync; clear the browser and it's gone.

license

MIT. The casebooks themselves belong to their schools; this repo only stores links to them.