Real-time voice AI agent
Sofia
A 24/7 phone receptionist for clinics: a speech→text→LLM→speech cascade with sub-1.5s median latency, idempotent tool-calling, and provider failover (Claude Haiku 4.5 / Gemini 2.5 Flash).
- Python 3.12
- LiveKit Agents
- Deepgram Nova-3
- Claude Haiku 4.5
- Gemini 2.5 Flash
- ElevenLabs Flash
- Silero VAD
- PostgreSQL
- Docker









network/
Finished Kaggle’s Intro to Deep Learning — first neural nets, built and trained in Keras. Classical ML → deep learning.
Shipped Sofia’s real-time voice pipeline — Deepgram → Claude Haiku → ElevenLabs, ~1.4s median latency per turn.
Another Kaggle Learn certificate down — twelve now, and counting.
Re-riced Arch: Catppuccin Mocha everywhere, Colemak-DH on a split keeb.
Placed 4th in Valencia and 11th in Spain in competitive programming.
Grew Competitive Programming UPV from inactive to the largest CP team in Spain.