AI Interview Assistant

Real-time interview coach that transcribes live calls and generates context-aware answers from your resume and job description.

The Problem

Technical interviews are stressful. You're simultaneously listening, thinking, coding, and communicating — all while being evaluated. What if you had an AI co-pilot that could listen to the conversation and surface relevant answers from your own experience?

The Solution

AI Interview Assistant is a desktop overlay app that:

1. Listens to your interview audio in real-time (via virtual audio loopback)
2. Transcribes speech using Groq's Distil-Whisper (sub-2s latency)
3. Generates contextual answers using your resume, job description, and personal stories
4. Displays responses in a non-intrusive "Dynamic Island" style overlay

Features

• Real-time transcription — Groq Whisper processes audio chunks as they arrive
• Context-aware answers — grounded in your uploaded resume + JD + personal anecdotes
• Story Engine — RAG-based retrieval of your personal experiences via SentenceTransformer embeddings
• Dynamic Island UI — frameless, transparent PyQt6 overlay that expands when answers arrive
• Interview reports — post-call summary with all Q&A pairs saved to SQLite
• Multi-LLM fallback — ZhipuAI GLM-4 → OpenRouter Llama 3 → Groq, with automatic failover
• Hotkey controls — R to regenerate, customizable shortcuts
• Privacy-first — no audio stored, transcript-only persistence

Tech Stack

Component	Technology
UI Framework	PyQt6
Transcription	Groq Distil-Whisper
LLM (primary)	ZhipuAI GLM-4
LLM (fallback)	OpenRouter (Llama 3), Groq
Embeddings	SentenceTransformer (all-MiniLM-L6-v2)
Audio	sounddevice + webrtcvad
Database	SQLite
Resume parsing	pypdf

Quick Start

Prerequisites

• Python 3.10+
• A virtual audio loopback driver (BlackHole on macOS, VB-Cable on Windows)
• API keys: Groq, OpenRouter, and/or ZhipuAI

Installation

git clone https://github.com/nnish16/ai-interview-assistant.git
cd ai-interview-assistant
pip install -r requirements.txt

Running

python main.py

The setup wizard will guide you through audio device selection and API key configuration on first run.

Audio Setup

To capture interviewer audio from Zoom/Teams/Meet:

1. Install a virtual audio loopback (e.g., BlackHole 2ch)
2. Create a Multi-Output Device in Audio MIDI Setup (macOS) combining your speakers + BlackHole
3. Set the Multi-Output Device as your system output
4. In the app, select BlackHole as the input device

Architecture

┌─────────────────────────────────────────────┐
│                  main.py                     │
│           (PyQt6 Application)                │
├─────────────┬───────────┬───────────────────┤
│  Audio      │  LLM      │  Story Engine     │
│  Service    │  Service   │  (RAG)            │
│  ┌────────┐ │ ┌────────┐│ ┌───────────────┐ │
│  │sounddev│ │ │ZhipuAI ││ │SentenceTransf.│ │
│  │webrtcvad│ │ │OpenRtr ││ │SQLite vectors │ │
│  │Groq    │ │ │Groq    ││ │Resume+JD ctx  │ │
│  └────────┘ │ └────────┘│ └───────────────┘ │
├─────────────┴───────────┴───────────────────┤
│          Dynamic Island UI (PyQt6)           │
│     Frameless • Transparent • Always-on-top  │
└─────────────────────────────────────────────┘

Roadmap

• Windows + Linux audio setup guides
• Streaming transcription (no chunking delay)
• Export interview prep deck (PDF)
• Browser extension for web-based interviews
• Test coverage > 80%

Contributing

See CONTRIBUTING.md for development setup and guidelines.

License

MIT — Nishant Sarang