Aurora Forge
Practice for the AI-assisted coding interview
10 warm-up questions. 8 real-round bugs. 16 coding problems from real interviews.
1 · Warm-up · 2 · Real-round bugs · 3 · Coding problems · Real questions
[!IMPORTANT]
Meta is rolling out, and Google is piloting, coding rounds with an AI assistant. You're graded on
judgment: scoping a vague ask, reading code you didn't write, catching what the model got wrong.
LeetCode doesn't train that. This does.
Three layers
| Layer | What it is |
|---|
| 1 | Warm-up | 10 hard multiple-choice questions on how AI coding tools behave. Finishing sends you to Layer 2. |
| 2 | Real-round bugs | 8 bugs from reported AI-enabled rounds, Easy to Hard. Find the cause; catch the AI's wrong fix. |
| 3 | Coding problems | 16 runnable problems, each a failing test suite plus a mock interviewer. 01, 06 and 12 run as a timed round: scope, phases with checkpoints, curveball, scorecard. |
Every problem is tagged with where it was asked. The question bank
lists every reported theme with its source (docs/reported.js; how it's built:
REPORTED_PROBLEMS.md).
Problems
Quickstart
git clone https://github.com/kenny2077/Aurora-Forge.git && cd Aurora-Forge
cd problems/07_Concurrency_Race && python3 -m pytest -q # Python: watch it fail, fix it with AI
cd problems/05_Rate_Limiter_TS && node --test # TypeScript: Node 24+, no build
python3 interview.py 01_Line_Wrap # mock interview in the terminal
Every starter fails on purpose. Make it green with an AI assistant, and never keep a line you can't
explain. Log prompts and checks in each problem's AI_LOG.md.
Layout
problems/NN_*/ README ticket, starter code, failing tests, interview.json (mock), AI_LOG.md
interview.py terminal mock interviewer
docs/ the site: index (Layer 1), swe (2), days (3), questions (bank), reported.js (data)
License
MIT.
