I recently built VoiceIQ — a complete Voice AI pipeline that listens to your voice, thinks using an LLM, and speaks the answer back. The best part? It costs nothing.
The Stack
- Whisper Large V3 (via Groq API) → Speech to Text.
- LLaMA 3.1 8B Instant (via Groq API) → Language Model.
- gTTS → Text to Speech.
- Streamlit → Web UI.
Feature: Conversation Memory
By default, every LLM call is stateless. I fixed this by building a ConversationMemory class that stores the last 8 turns and passes full history with every request.
Real Bug I Hit
Mid-development, Groq deprecated llama3-8b-8192. App threw a 400 error. Fix was simple — switch to llama-3.1-8b-instant. But it taught me to never hardcode model strings!
Why Groq over OpenAI?
Groq offers free tier with very fast inference. For a voice assistant, speed matters more than marginal accuracy gains.
Watch the Full Demo
I made a complete video walkthrough showing the live pipeline:
👉
Happy to discuss the implementation in comments!
Top comments (0)