A personal, compounding interview-prep system you run with an AI coding agent. You drop in raw material (mock-interview recordings, interview notes, project write-ups, articles, recruiter chats); the agent synthesizes it into reusable answers, frameworks, company briefs, and scored debriefs. Every interview you feed it makes the next prep sharper.

It's built on a three-layer "context engineering" pattern (popularized by Andrej Karpathy): keep the raw material immutable, let the agent maintain a wiki on top of it, and govern the whole thing with one operating-manual file.

This is a template. Click the green "Use this template" button at the top of the GitHub page to get your own copy, then replace the [placeholders] in CLAUDE.md / AGENTS.md.

How it works

sources/   → Layer 1: raw material you drop in. Immutable. You own it.
wiki/      → Layer 2: the agent's synthesized notes. The agent owns it.
CLAUDE.md  → Layer 3: the operating manual — the rules the agent follows.
AGENTS.md

The loop:

  1. You drop a file into sources/ (a transcript or interview notes, a project write-up, an article).
  2. You ask your agent to ingest it. The agent mines the durable content into wiki/ and moves the source into a processed/ subfolder.
  3. You query the agent; it answers from the wiki.
  4. After an interview, you feed in your notes — or, for a mock you recorded, the transcript; the agent scores your answers and tells you exactly what to improve.

Because the agent always updates existing notes before creating new ones, the wiki compounds instead of sprawling.

What the agent maintains (wiki/)

  • behavioral/ — one file per behavioral theme, each holding STAR stories from your real projects.
  • frameworks/ — reusable analytical templates (GTM, prioritization, pricing, etc.).
  • companies/<company>/ — company overviews and role-specific briefs.
  • ai-knowledge/ — domain knowledge worth remembering.
  • interview-debriefs/<role>/ — per-interview debriefs, scored against the role's rubric.

The exact rules for each live in CLAUDE.md / AGENTS.md. Read that file first.

Quick start

  1. Use this template → create your own repo. Keep it private (see Privacy below).
  2. Clone it to your machine.
  3. Point your AI coding agent at the folder (Claude Code, Cursor, etc.). It reads CLAUDE.md / AGENTS.md automatically.
  4. Open the [placeholders] in CLAUDE.md / AGENTS.md and set your target functions and companies.
  5. Drop your first project write-up into sources/projects/ and ask the agent to run ingest-project.
  6. (Optional) Open the folder as an Obsidian vault so [[wikilinks]] render natively.

Prerequisites

  • An AI coding agent — e.g. Claude Code, Cursor, or any agent that reads CLAUDE.md / AGENTS.md. (Required.)
  • Obsidian — optional, for rendered wikilinks and a graph view. The wiki is plain markdown, so any editor works.
  • Local transcription — optional, only needed to turn mock-interview audio recordings into transcripts:
    • openai-whisper — runs locally, so your audio never leaves your machine.
    • ffmpeg on your PATH.
    • The exact command is in the Tooling section of CLAUDE.md.

Privacy & safety — read before going public

As you use this, sources/, wiki/, log.md, and index.md fill up with real, sensitive content: company names, interviewer names, confidential project details, and your own candid self-assessments.

  • sources/ is git-ignored by default (raw, private, large). Only the empty folder skeleton is tracked.
  • wiki/, log.md, and index.md are tracked so you get version history — which is great in a private repo.
  • Keep your working repo private. If you ever want to share your setup publicly, share this template, not your filled-in repo.
  • The .gitignore has a clearly marked block you can uncomment to exclude your wiki from git entirely.
  • Never commit a .env or any secrets.

Customizing

  • CLAUDE.md and AGENTS.md ship with identical content so the system works with any agent (Claude Code reads CLAUDE.md; many others read AGENTS.md). If you change the rules, edit both — or replace one with a one-line pointer to the other to avoid drift.
  • Define your debrief scoring rubric the first time you run a debrief (see "Open items" in the manual).

Folder structure

sources/
  recordings/        (+ processed/)   mock-interview audio/video — agent transcribes
  transcripts/       (+ processed/)   verbatim text from those recordings
  interview-notes/   (+ processed/)   real-interview or recruiter/advisor notes
  reading/           (+ processed/)   articles, chapters, news
  projects/          (+ processed/)   your detailed project write-ups
wiki/
  companies/
  behavioral/
  frameworks/
  ai-knowledge/
  interview-debriefs/
CLAUDE.md / AGENTS.md            the operating manual
index.md                        catalog of every wiki page
log.md                          chronological record of changes