Quick answer: you don’t need another system design course — you need your existing material to be searchable. The primer repo, your own notes, and the mock interviews you’ve recorded can become one local library you can search by keyword or meaning while you practice, for free.

This is not a system design guide. DesignGurus, ByteByteGo, and the System Design Primer teach the material better than a search-tool blog ever will. This post is about the workflow around that material: collecting it, searching it, and — the part almost nobody does — reviewing your own mock interviews as text.

Disclosure: I build StashBase, the tool used below. Everything in the library method works with your own files and openly licensed material. Current as of August 2026.

What goes in the library

  • System Design Primer — the 300k-star study guide is licensed CC BY 4.0, so you can clone it straight into your library.
  • Your own notes — Markdown, docs, diagrams, screenshots of whiteboards. Scattered formats are fine; that’s the point of indexing them.
  • Your mock interview recordings — record your practice sessions (with your partner’s consent), and drop the audio files in. This is your highest value corpus and only you have it.
  • Books you own — buy them. Pirated PDF collections aren’t worth the habit, and this workflow doesn’t need them: the openly licensed material plus your own output is already more than you’ll finish.

If you want more than the primer, we measured GitHub knowledge bases worth cloning — including four more interview repositories — with file counts, sizes, and licenses for each one.

Step 1: Put it in one folder

mkdir ~/sd-prep && cd ~/sd-prep
git clone https://github.com/donnemartin/system-design-primer
mkdir notes mocks

Move your existing notes into notes/, recordings into mocks/. No reorganizing needed beyond that — search replaces filing.

Step 2: Open the folder in StashBase

Install StashBase, open ~/sd-prep, and let it prepare the folder: Markdown and PDFs get extracted text, images get OCR, and audio files get timestamped transcripts from a local Whisper model — no API key, no upload, no per-minute fees. A two-hour mock costs nothing to transcribe; it just takes machine time.

Step 3: Read your mocks as text

This is the step that changes prep quality. A recording you’ll never re-listen to becomes a transcript you can search across sessions:

  • “Every answer where I brought up sharding” — did you say the same shallow thing three mocks in a row?
  • “back-of-envelope” — find every capacity estimate you attempted and check them against your notes.
  • Timestamps take you back to the exact moment in the audio when you want to hear how you actually sounded.

Step 4: Search while you practice

Keyword search works out of the box and is free. For semantic search, add your own OpenAI or OpenRouter key — then “how do I handle hot partitions” in your own words lands on the primer section plus your note from two weeks ago, even when neither says “hot partitions.”

Since the library is exposed over MCP, the agent you already use — Claude, Codex, Cursor — can search it too: “quiz me from my weakest topics based on my mock transcripts.” Setup for each client is in the docs.

The honest limits

  • StashBase is Alpha — expect rough edges.
  • Search by meaning needs your own OpenAI or OpenRouter key. Keyword search does not.
  • Local transcription speed depends on your hardware; long recordings take a while on modest laptops. Queue them overnight.
  • None of this replaces doing the mocks. It compounds them.