At Bigscal Technologies, I build production agentic platforms and multi-agent systems serving 1M+ active ERP users across 120+ enterprise tenants on Azure and AWS.
I architected LangGraph and SAP Joule 2.0 routing for 650+ dynamic MCP capabilities, then reworked execution around a capability-contract layer and unified ReAct loop. The result was 3× throughput, 70% fewer tokens, and 5× inference-cost efficiency.
I fine-tune and serve Qwen2.5-7B and E5-large-v2 on NVIDIA GPUs, and built a hierarchical RAG pipeline indexing 10,000+ technical documents with 90%+ Recall@10. I also created Prompt Lab to capture reasoning trajectories and human preference annotations for scalable RLHF workflows.
I'm also a vLLM contributor, with merged work improving XML tool parsing and multimodal decoder safeguards. My foundation spans Python backend engineering, cloud-native infrastructure, evaluation, retrieval, and healthcare AI pipelines.
