At Comprinno Technologies, I shipped an agentic Statement of Work drafting platform on Claude and Amazon Bedrock, cutting client document prep time 40%. I also deployed a real-time speech-to-speech voicebot with sub-500 ms latency and authored CloudFormation infrastructure that reduced environment deployment time.
At the UMass Amherst AI Lab, I built Python pipelines that turn raw LiDAR PCAPs into structured 3D point-cloud frames and annotations for autonomous-driving perception models. I also developed evaluation tooling to validate 3D cuboid labels and improve training-data quality.
For my Enterprise Agentic RAG project, I built a LangGraph ReAct workflow on Amazon Bedrock with a groundedness verifier that checks claims against cited chunks. Hybrid retrieval improved Recall@10 from 0.68 to 0.89.
I’ve also built a Strands-based research agent and a grounding stage for AI-generated code, plus MCP servers and an agentic analytics lakehouse on Bedrock AgentCore. My IoT monitoring platform sustained roughly 928 requests per second at p95 latency of 10.8 ms with zero failed checks.
India