What it is
Interviewer is a proof-of-concept desktop app for voice-enabled mock interviews. It generates interviewer personas from a real job description and your resume, runs a multi-phase session, and then writes feedback you can actually use.
The privacy rule is the product: it runs on your hardware. No data leaves the machine, no API keys, no subscription. Speech-to-text, local LLM reasoning, and speech-out all sit on top of Lemonade Server.
Why it matters
Most interview coaches want your resume in the cloud. I built this during my AMD Lemonade Team co-op as a local-first alternative you can run on a client device — including a box like the Strix Halo setup the README documents, with Qwen3-Coder-30B-A3B-Instruct-GGUF as the recommended model.
Dynamic personas and a graded, multi-phase flow were the point. A static question list is easy. A session that stays coherent through greeting, core questions, and wrap-up, on a local model, is the hard part.
How it works
The shell is Electron. The UI is React. A unified prompt system drives five interview phases, persona generation, document parsing (resume and job description), and structured feedback extraction.
The flow the README diagrams is straightforward: greeting, then the question sequence, then close — with an audio check and a self-intro along the way. Persistence is local JSON.
I contributed substantially to this repo (career-database attribution: commits on lemonade-sdk/interviewer through mid-2026, including Linux/Electron fixes, model-name migrations, and README structure).
Status
Open source under MIT on lemonade-sdk/interviewer. Community contributions are welcome. Cross-platform: Windows, macOS, and Linux.
Related
- Lemonade SDK — the local inference server it talks to
- Universal Conditional Logic — the Interviewer README points at arXiv:2601.00880
- Lemonade Marketplace — I added the Interviewer listing and icon there