Live Interview Agent

🎤 A one-PC Python agent that helps you handle AI-based interviews.
It listens to the interviewer, transcribes with Whisper, generates concise English answers with GPT, and shows them in a floating always-on-top window.

✨ Features:

• Whisper STT + GPT answers in real time
• Always-on-top floating GUI
• VU meter (text + graphical) to confirm audio input
• Translation (transcript / answer / both) with hotkeys
• Chat timeline pane with save/clear options
• Optional TTS to hear answers in your ear
• Pro hotkeys: STAR mode, Concise vs Elaborate, copy, transparency, etc.

---

Quick Start

0) Prereqs

• Python 3.10+
• An OpenAI API key set as an env var:
  • Windows (PowerShell): setx OPENAI_API_KEY "sk-..." then restart terminal
  • macOS/Linux (bash): export OPENAI_API_KEY="sk-..."

1) Install dependencies

pip install -r requirements.txt

2) Route system audio to the agent

• Windows: Install VB-CABLE or use Voicemeeter Banana. Send your browser/app output (AI’s voice) to CABLE Input, and select CABLE Output as the agent input.
• macOS: Install BlackHole (2ch). Route browser output into BlackHole, set agent to record from it.

💡 Tip: Use headphones to avoid feedback.

3) Run

python live_interview_agent_v5.py --device-index <index_of_CABLE_Output>

4) During interview

• Keep the floating window visible.
• Read/paraphrase the on-screen answer naturally.
• The interviewer hears only your mic; the agent listens through the virtual cable.

---

Hotkeys

• F1: Toggle TTS (on/off)
• F2: Clear live fields
• F3: Pause/resume listening
• F4: Copy last answer to clipboard
• F5: Toggle STAR mode (Situation → Action → Result)
• F6: Toggle Concise vs Elaborate answers
• F7/F8: Adjust window transparency
• F9: Cycle translation mode (off / transcript / answer / both)
• F10: Change target language code
• F11: Save timeline to file (txt)
• F12: Clear timeline pane
• ESC: Quit

---

Configuration (via env vars)

• AGENT_CHUNK_SECONDS (default 2.5) — audio slice length
• AGENT_SAMPLE_RATE (default 16000) — capture rate
• AGENT_SILENCE_THRESHOLD (default 0.003) — RMS gate for silence
• AGENT_STT_MODEL (default whisper-1)
• AGENT_GPT_MODEL (default gpt-4o-mini)
• AGENT_SHOW_TRANSCRIPT (1 to display recognized question)
• AGENT_TTS_DEFAULT (1 to start with TTS enabled)
• AGENT_TRANSLATE (off|transcript|answer|both)
• AGENT_TARGET_LANG (default es)

Example (Windows PowerShell):

setx AGENT_SHOW_TRANSCRIPT "1"
setx AGENT_TRANSLATE "both"
setx AGENT_TARGET_LANG "fr"

---

Troubleshooting

• No VU activity → check that browser/app output is routed to CABLE Input.
• VU moves but no transcript → raise volume or lower AGENT_SILENCE_THRESHOLD.
• ImportError → pip install -r requirements.txt
• TTS issues → disable with F1 or set AGENT_TTS_DEFAULT=0.

---

Safety & Ethics

This tool is meant to support comprehension and confidence. Use responsibly and check the rules of your interview platform.