At Nyra AI, I built a production-grade voice-calling agent for hospitals that handles inbound and outbound patient calls. I designed a low-latency pipeline connecting speech recognition, LLM reasoning, and speech generation.
I developed scalable backend services with Python and FastAPI, using asynchronous processing and WebSockets for real-time communication. I also integrated SIP telephony and worked with LiveKit, Sarvam AI, OpenAI, and AWS deployment.
At Heyrik, I trained and fine-tuned open-source models for speech and generative AI applications. I extended a Text-to-Speech model with voice cloning and developed audio preprocessing and speaker-conditioning pipelines.
My projects include backend work for SignBuddy, Husle Life, FestGo, and Subhamastro. I also won a 24-hour inter-college hackathon with a code generator using local LLMs and placed second at the HydPy Hackathon with a voice-based agent.
