Interview Prep

Hands-on practice for junior interviews: backend, data analyst and AI engineer on one shared core, plus object-oriented design in practice and a reverse engineering sprinkle. Short narrated lessons, exercises with real tests, and a tutor that reads your code.

Created by Avi.

Requirements

• Node 24 or newer, and pnpm
• Python 3.13 with pip install -r requirements.txt (for the Python lessons)
• VS Code with its code command on your PATH (the trainer runs code serve-web)

Run it

pnpm install
pnpm start            # builds lessons, installs the extension, opens VS Code in the browser

VS Code opens with the next unfinished lesson on the right and your exercise file on the left. First time only: click Trust when VS Code asks (it is your own folder, served only to this machine, behind a token). The URL keeps the same token between runs, so you can bookmark it.

Press Run my tests in the lesson, or pnpm lesson test <track>/<lesson>. Pass and the next lesson opens. A run that never finishes (an infinite loop in your code) is stopped after 90 seconds with a hint.

Voice

pnpm narrate          # records the spoken script of every lesson with Edge Read Aloud voices, cached
pnpm build            # rebuilds pages with the audio

Every narrated card has a play button. Parts of each diagram light up as the voice reaches them.

The voices come from edge-tts-as-extension, a git submodule in vendor/. The first pnpm start or pnpm narrate checks it out and installs it, with no other setup. pnpm start also builds its recorder, so the voice in a lesson's narration panel can be switched to any Edge voice.

Tutors

• Claude Code: the Ask Claude button (or claude then /tutor) opens a Socratic coach that reads your lesson, your code and your progress. It gives hints in steps, reviews against the project style, and runs mock interviews
• Gemini Live: a voice tutor that watches your VS Code and talks you through it. Click Voice tutor in the VS Code status bar. It runs either on your own Gemini API key (no extension needed) or on your Gemini subscription through the apps/gemini-live Chrome extension. See Gemini Live

Gemini Live

The voice tutor talking through a lesson in real time:

2026-10-26.14-40-25.msedge.mp4

Warning

The subscription engine rides your Gemini subscription through Google's own web client, which Google doesn't offer for this use. Google may suspend or ban an account that uses it. If you'd rather not risk your account, use your own Gemini API key - the official, public Live API.

• Your API key (no extension): click Voice tutor in the status bar. The page that opens asks for a key once (create a free one in Google AI Studio) and keeps it in VS Code's secret storage. A GEMINI_API_KEY environment variable, if set, takes precedence
• Your Gemini subscription: sideload the extension (the same page walks you through it, or see its README). Once it is installed, its popup switches between the subscription and your API key

Layout

Path	What
curriculum/<track>/<NN-slug>/	lesson.md, starter/, solution/, tests/
curriculum/_kit/	helpers lessons share: implementation loading, Postgres (PGlite), fake LLM, mutation checks, seeded random, timing, bookstore Money
curriculum/LEARNER.md	who the course is written for
curriculum/AUTHORING.md	how to write a lesson (enforced by pnpm verify)
curriculum/tracks.json, roles.json, path.json	tracks, target roles and the study order
work/	your copies of the exercises (git-ignored)
.progress/	what you passed (git-ignored)
scripts/	CLI (pnpm lesson), build, narrate, verify, start; scripts/lib holds the shared helpers
apps/vscode-ext/	the VS Code extension
apps/gemini-live/	the Gemini Live bridge
vendor/edge-tts-as-extension/	the voice tool (git submodule)

Commands

pnpm lesson list | next | open <id> | test <id> | reset <id>
pnpm verify [track]        # every solution passes, every starter fails, structure rules hold
pnpm typecheck             # scripts, curriculum and every solution (run pnpm verify first)
pnpm build                 # lessons -> dist/

Why it is built this way

• Short cards (130 words max), a picture on each, a voice that walks it: the course is built for someone who loses the thread in long text
• Every lesson names the rejected alternative and its cost, because interviews ask "why not X?"
• Every exercise ships a failing starter and a passing solution, checked by pnpm verify, so a green test means something
• SQL runs on real Postgres in-process (PGlite), AI exercises use a scripted fake model, so nothing needs a key or a network

The default order puts SQL first, then tests and APIs, then each role's branch. Change it in curriculum/path.json.

Make it yours

The course is data in curriculum/, so you can add lessons, tracks and roles, or retarget it to another job or subject, without touching the tooling.

• With a coding agent: ask it to extend the curriculum. Claude Code has the /extend-curriculum skill, and other agents read AGENTS.md, which points to the same steps
• By hand: follow .claude/skills/extend-curriculum/SKILL.md, then curriculum/AUTHORING.md. pnpm verify tells you when a lesson is right

License

Apache License 2.0. Copyright 2026 Avi.

You may use, change and share this project, including commercially. If you fork it or redistribute it, keep the NOTICE file and its credit to the original author, as section 4(d) of the license requires, and mark the files you changed.