Back to Portfolio
Shipped Open Source Asset August 2026 11 min read

TakeMyInterview AI: Offline Voice-to-Voice AI Job Interviewer Platform

Open-source voice-to-voice interview practice application engineered with Whisper C++ speech recognition, quantized local Llama 3.3 LLM scoring, and sub-100ms Piper neural TTS synthesis running 100% offline.

GJS
Lead AI Architect & CS FounderPunjab, India • MSME Certified

Gurdharam Jeet Singh

Founder & Lead AI Architect at Gurdharam AI Engineering. Leading a CS engineering squad building direct Meta Cloud API WhatsApp bots, Indic Voice AI calling agents (Sarvam AI), edge computer vision models (TFLite/Flutter), and 3D WebGL platforms for clients globally.

Meta Cloud API Edge AI / TFLite OWASP Security Audited

System Performance Benchmarks

100% Offline
Zero Cloud Data
< 320 ms
Audio Latency
STAR Model
Behavioral Scoring
$0 / Mo
API Cost

1. The Need for Privacy-First Voice Interview AI

Job applicants and HR talent platforms preparing for technical software engineering, product management, and executive interviews face high friction with commercial AI tools. Existing platforms transmit private voice recordings, resumes, and confidential interview answers to cloud server logs.

TakeMyInterview AI was developed as an open-source, air-gapped desktop application where users practice real-time spoken technical interviews without internet connectivity or subscription fees.

2. Voice-to-Voice Architecture Diagram

// TakeMyInterview Voice-to-Voice Pipeline
[Microphone Audio Stream (16kHz PCM)]
├──> 1. Whisper C++ Speech-to-Text (On-device transcription in ~90ms)
├──> 2. Local Llama 3.3 70B / Qwen 2.5 LLM (STAR Scoring & Follow-up Question)
├──> 3. Piper Neural Text-to-Speech (Streaming audio synthesis in ~80ms)
└──> [Speaker Output: Live Conversational Interruption & Feedback Dashboard]

3. Local Audio Stream & Scoring Setup

// Local Node.js Audio Processing & Local LLM Scoring Pipeline
const { WhisperCpp } = require('whisper-node');
const { PiperTTS } = require('piper-tts-node');

// 1. Air-gapped on-device speech transcription
const whisper = new WhisperCpp({
  modelPath: './models/ggml-base.en.bin',
  whisperCppPath: './bin/whisper'
});

// 2. Real-time interview evaluation callback
async function processCandidateAnswer(pcmBuffer) {
  const text = await whisper.transcribe(pcmBuffer);
  
  // Local vLLM STAR scoring
  const response = await fetch('http://localhost:8000/v1/chat/completions', {
    method: 'POST',
    headers: { 'Content-Type': 'application/json' },
    body: JSON.stringify({
      model: 'Qwen2.5-70B-Instruct',
      messages: [
        { role: 'system', content: 'Evaluate candidate interview answer using STAR model format.' },
        { role: 'user', content: text }
      ]
    })
  });

  const data = await response.json();
  return data.choices[0].message.content;
}

Frequently Asked Questions

Is TakeMyInterview AI free and open-source on GitHub?

Yes. The source code is publicly hosted at github.com/gurination1/TakeMyInterview-Ai for developers and job seekers.

Can enterprise HR platforms license this voice AI interviewer engine for white-label hiring automation?

Yes. We deploy custom white-label instances integrated with enterprise Applicant Tracking Systems (ATS) and internal candidate scoring pipelines.

Build White-Label AI Interviewer Platforms

Build custom voice-to-voice candidate screening engines and automated technical interviewing platforms.

Consult AI Voice Architect