I tried building a voice bot that answers questions about my resume and projects out loud. First, I wrote my resume and project details into files the bot could search through. Then I set up accounts for the voice tools like LiveKit, Deepgram, and ElevenLabs and planned to use Anthropic's Claude for the bot's brain, but it needed a payment method I didn't have. So I switched to Google Gemini's free API instead, and updated my code to use it. Next, I installed Python and all the required packages in VS Code, but my first attempt failed because my project files were nested inside an extra folder ,I found and fixed that. When I finally ran the bot, it connected properly and loaded all four voice/AI plugins, but gave a 401 error, meaning my LiveKit key didn't match.
Voice pipeline (LiveKit Agents framework, Python):
STT → Deepgram
LLM → Gemini AI with a RAG layer over source docs
TTS → ElevenLabs
Turn detection / VAD → LiveKit's built-in