biq

A behavioral interview question bank with hire / no-hire example sheets at Junior, Senior, and Exec.

App card

This repo ships card.json and icon.png as the listing for any host (kindel.com, iOS, Android).

Live site: https://kindel.com/kld/apps/biq/. Amazon is the default. Arm is https://kindel.com/kld/apps/biq/?c=arm.

The bank is company-scoped. data/questions.json has a companies array. Each company has its own principles and questions. The picker writes ?c=arm (or drops the param for Amazon) so a link without c stays Amazon.

Run

Needs a static file server because the bank is loaded with fetch.

python3 -m http.server

Open http://127.0.0.1:8000/

Host configuration

The default URLs are root-relative, which assumes the site is served from /. A host that mounts biq somewhere else can override them by setting window.BIQ in an inline script before js/biq.js or js/biq-examples.js loads:

<script>
  window.BIQ = {
    questions: "/biq/data/questions.json",   // question bank
    examplesData: "/biq/data/examples/",     // per-question example packs
    examplesFallback: "/biq/data/biq-examples.json", // optional legacy single-file bank
    examplesPage: "examples.html"            // resolved against the current page
  };
</script>

Check the search

node scripts/check_search.js

Lifts norm and matches out of js/biq.js rather than restating them, so the check cannot drift from what the page runs. It asserts that every name and alias finds its own principle, that junk and stop words match nothing, that each abbreviation resolves to the one principle it abbreviates, and that near misses land.

Regenerate examples

Run the Generate example packs action. It is manual only, because every real run spends xAI credits. Pick a company, leave dry_run on for the first pass to see how many calls it would make, then run it again with dry_run off. The packs arrive as a pull request rather than on the default branch, because they are model-written interview content and a human should read a few first.

The key lives in the XAI_API_KEY repository secret. There is no reason to hold it locally.

Resume-safe. It writes data/examples/{slug}.json only for a question that does not already have a valid junior / senior / exec pack, so re-running after a partial failure costs only what is still missing.

Locally, for a count without a key:

BIQ_DRY_RUN=1 python3 scripts/generate.py
BIQ_DRY_RUN=1 BIQ_COMPANY=gitlab python3 scripts/generate.py

BIQ_COMPANY scopes a run to one company. Generating does not turn the buttons on: "examples": true in data/questions.json is a separate, deliberate edit.

The slug is an 8-character FNV-1a of str(principle_id)|norm(question). The numeric principle id is unique across all companies (Amazon 1000s, Arm 2000s, Coupang 3000s, Delivery Hero 4000s, GitLab 5000s, Dawn 6000s), so pack filenames cannot collide even when two companies share a principle name. The generator reads prompt.md and calls xAI grok-4.3.

License

MIT. Copyright (c) 2026 Kindel, LLC. Keep the copyright notice and permission notice in all copies.