seemanth raju Kurapati

Python, FastAPI. Realtime voice: Pipecat, LiveKit, WebRTC, streaming ASR→LLM→TTS. PyTorch, TensorFlow, LangChain, RAG. Postgres, Docker, GCP, Azure. Specialized in voice, generalist backend.

Flow-matching TTS decoders and low-NFE inference. Triton and vLLM-style model serving. Formal speech eval WER, MOS, PESQ, UTMOS. Filling gaps my production work never forced me into.

A harness that timestamps every stage of a voice turn end-of-speech, ASR, LLM first token, TTS first byte and plots the tail, not the average.

Talking voice agent latency, turn-taking and the failures that only show up on real phone calls. Also open to AI engineering roles realtime voice or agentic backends.