At HCLTech, I architected an enterprise GenAI solution for SDLC automation, using Supervisor-Worker agents and shared OpenSearch RAG. The workflows generated and published requirements and design documents across Confluence and Jira, reaching approximately 90% PDM adoption for requirements workflows.
I also built Jira and Confluence API agent tooling and fine-tuned models for sensitive-data detection and synthetic Q&A generation. My DeBERTa-v3 PII detection model achieved an F1 score of 96% on 200,000+ samples, and I exported a fine-tuned Phi-3 Mini model to ONNX for edge inference.
At Samsung Semiconductor India Research, I engineered a tiered memory caching allocator in PyTorch’s C++ backend, reducing allocation latency by 15% and improving DLRM throughput by 5%. I also built TraceIQ, a local-first document traceability agent, and have two published research papers at IEEE-indexed conferences.
