# Zizo's Team

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Zohaib Usman](https://cerebralvalley.ai/u/zizohere)
- **GitHub:** https://github.com/z1xO/voxora
- **Demo video:** https://youtu.be/seqyp_XBCxM
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2784

Voxora is a mock technical interview coach for computer science students. Most students practice coding problems in silence, but real interviews ask you to think out loud: explain your plan, take questions, and talk through trade-offs while you type. Voxora practices that part.

Pick a problem. Five curated Python problems (Two Sum, Valid Parentheses, Merge Intervals, Longest Substring Without Repeating Characters, Product of Array Except Self), each with a prompt, examples, constraints and starter code.
Solve it out loud. You code in a Monaco editor (the editor from VS Code) and talk. Your speech is transcribed in your browser by Whisper, and you can type instead or correct the transcript at any time.
A Llama interviewer follows along. It sees the problem, what you just said, and short summaries of how your code changed. It behaves like a real interviewer: it stays quiet while you work, answers your questions, asks one question at a time at natural moments ("walk me through your approach", "what's the complexity?", "can you do better?"), and speaks in a natural voice.
Hints, never answers. Three staged hints, from a nudge to a near-explicit step, capped by hints written for each problem. If you ask for the solution, it kindly declines and steers you back.
Run the tests. Sample tests run in a locked-down Docker sandbox (no network, 128 MB of memory, read-only, non-root), or in your own browser with Python compiled to WebAssembly when no sandbox is available, as on the hosted site.
Get an evidence-based scorecard. Problem solving, communication, complexity analysis and edge cases are each scored 1 to 5. Every score must cite a real quote from your transcript, a line of your code, or a test result, and Voxora checks each citation against the session. A score it can't verify becomes "insufficient evidence" instead of a guess. You also get two strengths and up to three concrete next steps.
Private by default. With the local model, your transcript and code never leave your computer. Raw audio is never stored, and saving a session is optional and deletable.

## More from Meta Global AI Developer Hackathon

- [SpeakEasy](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2781)
- [Md. Fatin Shadab Turja](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2782)
- [Arnav.S*k-3](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2783)
- [Vist](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2785)
- [SciSleuth AI](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2786)
- [Tiny Libraries](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2787)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2784. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt