🎙️ SDET Interview Coach

A fully local, privacy-preserving, terminal-based AI application designed to help you practice for SDET (Software Development Engineer in Test) interviews using voice-to-voice interaction.

Powered by Whisper (Speech-to-Text), Piper (Text-to-Speech), and Ollama (Local LLMs like Llama 3.1 & Qwen).

---

🚀 Key Features

• Local & Private: No audio or text data ever leaves your machine.
• Voice-to-Voice Interaction: Practice speaking your answers just like a real interview.
• RAG (Retrieval-Augmented Generation): Ingest your own study materials (.pdf, .txt, .md) to ground the AI's questions in your specific resources.
• Multi-Round Support: Practice specifically for HR, Technical, Managerial, or CTO rounds.
• Session Intelligence:
  • Real-time timer and progress tracking.
  • Active keyboard overrides (done, quit, skip) while the mic is live.
  • Smart handling of strict model templates (optimized for Llama 3.1 & Qwen).
• Post-Interview Report: Get an immediate JSON-formatted feedback report with scores, strengths, and areas for improvement.

---

🛠️ Tech Stack

• Ollama: Orchestrates local Large Language Models.
• Faster-Whisper: High-performance local speech-to-text.
• Piper TTS: Low-latency, ultra-fast local text-to-speech.
• Sentence-Transformers: Local embeddings for the document knowledge base.
• Rich: Beautiful terminal UI with progress bars and panels.

---

📦 Installation & Setup

1. Prerequisites

• Python 3.10+
• Ollama: Download and install
• FFmpeg: Required for audio processing.

2. Clone and Setup Environment

git clone <repository-url>
cd sdet-interview-app

# Run the automated setup script
chmod +x setup.sh
./setup.sh

The setup script will create a virtual environment, install dependencies, download the Piper binary, and fetch the default voice model.

3. Pull LLM Models

For the best experience, pull these models via Ollama:

# Recommended: Fast & Snappy
ollama pull llama3.1:8b

# Advanced: High Intelligence (Slow)
ollama pull qwen3:30b

---

📖 How to Use

Step 1: Index your study docs (Optional)

Place your .pdf, .md, or .txt files into resources/<topic_name>/ and run:

python main.py --index

Step 2: Start an Interview

python main.py
=======
./run.sh --index

Step 2: Start an Interview

The easiest way to start the app is to use the provided runner script which handles the virtual environment for you:

./run.sh

Manual Alternative (Using venv directly)

If you prefer to run it manually:

source venv/bin/activate
python3 main.py

Follow the interactive wizard to select your Model, Round, Topic, and Duration.

Step 3: During the Interview

When the 🎙 Recording… status appears:

• Speak your answer naturally.
• Silence Detection: Stay silent for 7 seconds (configurable) to auto-submit.
• Keyboard Overrides: Type these into the terminal while recording:
  • done + Enter: Instantly submit your current answer.
  • skip + Enter: Move to the next question.
  • quit + Enter: Gracefully end the interview early.

---

🔧 Troubleshooting & Performance

"Thinking" Time

If the AI takes too long to respond (the "🤖 Thinking..." spinner), it is likely because the model is generating internal reasoning tokens.

• Fix: Use a smaller model like llama3.1:8b.
• Note: We have implemented a specialized "thought stripper" that hides <think> tags from being displayed or spoken, ensuring you only hear the interviewer's actual response.

Empty Responses (Llama 3.1)

Llama 3 models are strict and require a user kickoff. We have optimized the InterviewSession logic to ensure the chat template is always correctly triggered, so the interviewer never stays silent.

Microphone Issues

• Ensure your default system microphone is correctly selected.
• If you see a PortAudio error, ensure you have ffmpeg and the sounddevice dependencies installed correctly via ./setup.sh.

---

📜 License

This project is open-source. Build, modify, and ace your interviews! 🚀

🎙️ SDET Interview Coach Walkthrough This document provides a comprehensive overview of the SDET Interview Coach, a private, local AI application for practicing interviews.

🚀 Features & Capabilities 100% Local: No data sent to the cloud. Voice-to-Voice: Interactive oral practice. RAG-Powered: Grounds questions in your own study documents. Feedback Engine: Generates professional scoring and recommendations. 🛠️ Setup Guide Dependencies: Run ./setup.sh to install everything automatically. Ollama: Ensure Ollama is running (ollama serve). Models: For high speed, use ollama pull llama3.1:8b. 📖 Using the App Starting a Session Run python main.py and follow the menu to choose your round (HR, Technical, etc.) and topic.

During the Interview The app uses 7-second silence detection to know when you've finished speaking. Type done: Instantly submit your answer without waiting for silence. Type skip: Move to the next question. Type quit: End the session and get your feedback report. 🔧 Optimized for Llama 3.1 & Qwen We've implemented specialized fixes to make the app work perfectly with the latest local models:

Thought Stripping: Automatically removes internal reasoning () tags so you only hear the final answer. Chat Template Fix: Uses a hidden "kickoff" message to ensure Llama 3 models always start the conversation properly. Context Reordering: Keeps system messages behind the user turn to prevent "silent generation" bugs in strict models. Good luck with your interview practice!

=======