Otofi — AI-Powered Real-Time Voice Calling Platform

Built the backend and real-time voice pipeline for Otofi (otofi.io), an AI-powered outbound voice calling platform. Designed the WebSocket/real-time layer in Python (FastAPI) integrating Twilio Media Streams with Deepgram's Voice Agent API for speech-to-text, LLM inference, and text-to-speech.

Key contributions:

Built a RAG/semantic search pipeline (Pinecone + BAAI/bge-small-en-v1.5 embeddings) for contextual response grounding

Integrated Deepgram Flux for accurate turn detection and reduced backchannel interruptions

Optimized latency (eager end-of-turn thresholds, speculative prewarm) and handled state reconciliation after user barge-in

Implemented human hand-off logic and Redis-based rate limiting (Upstash)

Shipped a live Zapier integration for the platform, now public in Zapier's app directory