WarmUp: A Private Voice Mock Interviewer Built for My Friend

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend

What I Built

I built WarmUp, an open-source, private voice mock-interview simulator, specifically for my friend Ayesha.

Ayesha has spent the last several months transitioning into software development and is currently preparing for technical screenings and behavioral rounds. While she knows data structures and full-stack development well, she struggles with interview anxiety—particularly framing answers under pressure using the STAR method (Situation, Task, Action, Result). Traditional human mock interviews felt intimidating, and paid coaching platforms were out of budget.

WarmUp provides an empathetic, pressure-free testing sandbox right on her laptop. She can speak her answers naturally, get real-time voice follow-ups tailored to her responses from multiple persona types (a supportive HR recruiter or a technical interviewer), and receive actionable feedback on clarity and technical depth.

Demo

• Live Application: [Deploy your app on Render/Vercel and paste link here]
• Video Walkthrough: [Add your short Loom or YouTube demo link here]

Code

The complete source code is available on GitHub:
👉 Naeemk10/warmup-interview-prep

The full codebase is hosted on GitHub under the MIT License, complete with setup instructions for local Ollama inference and the Next.js frontend.

How I Built It

WarmUp is built around local open-weight AI to provide low latency, zero API costs, and total privacy:

• Core LLM Engine: Google's Gemma 2 (2B-Instruct) running locally via Ollama. Gemma 2 was chosen for its precise instruction-following capabilities when roleplaying structured interviewer personas without requiring heavy compute resources.
• Roleplay Harnesses:
  • HR Persona: Listens for STAR framework elements, validates the answer warmly, and asks an organic behavioral follow-up (kept strictly under 50 words).
  • Technical Persona: Probes edge cases, time complexity, and architecture trade-offs.
• Voice Interface: Native Web Speech API for real-time speech recognition and synthesis, ensuring the app works out of the box with zero external speech API tokens.
• Frontend: Next.js (App Router) and Tailwind CSS with responsive visual states for active listening and speaking.

Why Does Open Innovation Matter?

Open-source and open-weight models are the only foundation that makes sense for an interview practice tool:

1. Complete Psychological Safety: An interview candidate should feel free to freeze, stutter, restart, and deliver rough practice answers without those transcripts being uploaded to commercial servers for model training. Everything stays strictly on the local machine.
2. Zero Practice Ceiling: Voice conversations consume massive amounts of tokens. Running open models locally removes the paywall completely, allowing Ayesha to practice for hours daily at zero cost.
3. No Vendor Lock-in or Model Drift: Running Gemma 2 locally ensures predictable system prompt responses without unexpected changes to safety filters or corporate API rate limits.

Prize Categories

• Best Use of Gemma (Powered by Gemma 2 via Ollama)