At Juspay Technologies, I build real-time conversational AI voice agents on LiveKit and WebRTC that transcribe, respond, and execute agentic actions during live calls. The system handles around 200 calls per day, with transcripts persisted to GCS at sub-0.5% data loss.
I re-architected speech-to-text from batch REST to real-time gRPC streaming, reducing latency from three seconds to under 500ms. I also own deployment and serving infrastructure for Silero VAD, transformer-based turn detection, and STT/TTS models, while building post-call workflows that generate summaries, tickets, and PRDs across 1,200+ jobs daily.
I also co-developed Claw, an agentic developer platform used by 90%+ of the team, and built reinforcement-learning-driven projects including a dark web crawler indexing 50,000+ pages at 92% classification accuracy.
