interview-coach

Learning project: a LangGraph interview coach and study buddy with measured RAG, long-term memory and evaluation at every level

Development

This project is managed with uv on Python 3.13.

Setup

git clone https://github.com/lbildzinkas/interview-coach.git
cd interview-coach
uv sync

uv sync creates a local .venv (git-ignored) and installs the project plus the dev tools — ruff, pyright and pytest — using the committed uv.lock so everyone gets the same versions.

Asking the coach a question

coach ask sends one question through a one-node LangGraph graph that calls GLM-5.3 on the GLM Coding Plan's OpenAI-compatible endpoint and prints the answer. Configuration lives in environment variables, loaded from a git-ignored .env file (copy .env.example to .env and fill it in; a real environment variable wins over the file):

• INTERVIEW_COACH_API_KEY — required, your GLM Coding Plan key
• INTERVIEW_COACH_BASE_URL — optional, defaults to the coding endpoint on open.bigmodel.cn
• INTERVIEW_COACH_MODEL — optional, defaults to glm-5.3

uv run coach ask "What is consistent hashing?"

Without a key, the command prints a readable error instead of a stack trace.

Looking at QASPER

coach qasper stats downloads the dev split of QASPER (questions about NLP papers, CC BY 4.0) into the git-ignored data/qasper/ folder, maps each answer's evidence to paragraph ids and prints how many questions fall in each category. No model is called and no API key is needed.

uv run coach qasper stats

Checks

One command runs every check that CI also runs — lint and formatting (ruff), type checks (pyright) and tests (pytest):

./scripts/check.sh

CI

.github/workflows/ci.yml runs the same command on every pull request and on pushes to master (GitHub Actions docs).

Local-only data

Three folders hold data that never leaves your machine; their contents are git-ignored from the start:

• data/personal/ — personal data (your notes, transcripts, answers)
• data/study-material/ — downloaded study material (guides and pages to review, Anki decks)
• data/qasper/ — the downloaded QASPER dataset cache

Downloading the study material

coach setup downloads the study material named in sources.yaml into the git-ignored data/study-material/ folder, each file pinned to a commit. It is safe to run twice: sources already downloaded are skipped. Credits are in THIRD_PARTY_NOTICES.md.

uv run coach setup

License

MIT — see LICENSE.