VirtualMentor: The AI Voice Coach That Takes Your Interview Prep Outdoors published: true

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass

What I Built

Preparing for technical and behavioral interviews often means being chained to a desk for 8+ hours a day—staring at problem sets, doomscrolling forum threads, and burning out in isolation.

VirtualMentor turns interview preparation inside out. Instead of another static coding test on a bright screen, it acts as an audio-first conversational sparring partner designed specifically for hands-free, outdoor walks. By untethering job seekers from their desks, VirtualMentor lets engineers and students practice system design explanations, behavioral storytelling (STAR method), and rapid technical questions while literally touching grass.

Who it's for:

• Students and software engineers preparing for tech interviews who suffer from screen fatigue.
• Candidates who need to build conversational confidence, verbal pacing, and articulate spoken delivery rather than just silent typing.

Demo

• *Live App / Web Demo:(https://virtualmentor.com)

Code

The project is fully open-source. Check out the repository for the full-stack codebase, API routes, and prompt pipelines: *(https://github.com/AnuragYadav9219/pitchPilot)

How I Built It

VirtualMentor is built around an open-source, local/self-hosted AI pipeline prioritizing low latency, natural conversational cadence, and privacy:

• Open-Source LLM Inference: Powered by open-weight models (such as Llama 3.1 8B Instruct / Mistral 7B) served via Ollama / vLLM, fine-tuned with structured prompts to act as a supportive yet rigorous technical hiring manager.
• Voice & Speech Processing:
  • Whisper (faster-whisper): Open-source speech-to-text running efficiently to capture candidate responses cleanly, even with ambient park/outdoor wind noise.
  • Open-Source TTS: Utilizes Piper / Kokoro TTS for sub-200ms latency, delivering responsive conversational turns without awkward pauses during a walk.
• Frontend & Real-time Flow: Built with React (Tailwind CSS, clean mobile-first viewport) connected to a real-time WebSocket backend, featuring a minimalist "pocket mode" that locks the screen and activates solely on voice activity.

Why Does Open Innovation Matter?

1. Accessibility for Every Candidate: Closed commercial APIs charge heavy per-token and per-minute audio costs that quickly price out students and unemployed job seekers. Open weights and local inference make high-repetition verbal drills practically free to run.

2. Privacy on Sensitive Career Data: Candidates discuss personal stories, career gaps, failures, and previous proprietary projects. Open-source deployment guarantees user audio and transcripts never feed proprietary training loops.

3. Modularity & Offline Independence: Open weights allow local quantization, enabling developers to run localized interview engines on-device without tethering to constant cloud API uptime.

Prize Categories

• Main Challenge: Week 1 — Touch Grass