I build real-time voice AI — the streaming STT → LLM → TTS pipelines behind phone agents that answer, understand, and talk back in under a second. Right now that means telephony for regulated BFSI and healthcare, on LiveKit + SIP.

now      Voice AI Engineer @ Ignosis AI       BFSI voice platform · sub-700ms · LiveKit SIP
before   Voice AI Engineer @ Core Cognitics   multi-tenant voice AI · 10k+ calls/day

~/ignosis  — A real-time voice platform for BFSI (debt-collection) telephony on LiveKit SIP (Exotel/Vobiz trunks). Targeting sub-700 ms end-to-end latency with full barge-in, per-provider STT/LLM/TTS failover, and an offline eval harness (WER + LLM-as-judge) that replays production calls to catch regressions before users do.

~/core-cognitics  — Co-led a multi-tenant voice AI platform on the LiveKit Agents SDK handling 10k+ calls/day across healthcare & automotive. Multilingual (Arabic / Indic / European) with mid-call provider switching, a multi-layer guardrail system, MCP tool servers, and a RAG stack on Qdrant.

~/open-source  — Found and reported a bug in the LiveKit Agents SDK (#5665) — STT errors closing the session without the tolerance counter LLM/TTS errors get.

Voice AI    LiveKit · Pipecat · Deepgram · ElevenLabs · OpenAI Realtime · Gemini Live · Silero VAD

LLM & Agents    Claude · OpenAI · LangGraph · MCP · RAG · Qdrant

Languages    Python · TypeScript · Java · SQL · Bash

Infra & Obs    Docker · Kubernetes · AWS · PostgreSQL · Redis · Kafka · Grafana · Prometheus · Langfuse

oneNote_AI  — Multi-agent model for academic note generation. _Python

AI-Call-Operator  — Multilingual voice call-center for disaster response — Sarvam TTS · Google STT · Llama 3. _Python

edulex · teacher  — AR adaptive-learning assistant for dyslexic students, voice-driven via Gemini Realtime. 🏆 ₹1L @ Spark Venture 2024. _TypeScript

Medical-chatbot  — RAG-powered medical assistant. _Python

_{Obsessed with latency budgets. Occasionally finishes side projects. 🎙️}