I build AI agents that do real work for businesses — agents that query databases, read documents, navigate knowledge graphs, and run multi-step workflows. Not demos. Systems built for production. Over the last few years I've shipped agentic AI across healthcare, clinical research, and BFSI: → A biomedical knowledge graph platform on Neo4j with an agent that answers drug-comparison and toxicity questions over CIViC data → LLM-powered clinical trial–patient matching on Snowflake → RAG systems for financial product documents, with chat over loan data and graph visualizations → Custom agent libraries, multi-model chat platforms (OpenAI, Bedrock, Gemini), and an LLM API proxy with usage tracking and budget controls What sets my work apart: I don't just use frameworks — I take them apart. I've reverse-engineered LangGraph, CrewAI, LangChain, and LlamaIndex to understand what's actually happening under the hood, then built lightweight custom agent harnesses that are faster, cheaper, and easier to control than the off-the-shelf stack. That includes deep work with MCP (Model Context Protocol), agentic RAG, human-in-the-loop interrupt patterns, and knowledge-graph-backed retrieval. Currently I'm a Lead Software Engineer in the AI Research team at Persistent Systems, where I design agent architectures and intelligent workflows that move GenAI from prototype to production. I also write about AI engineering — agent architectures, context engineering, RAG, and what actually works in production — for a growing audience of engineers on LinkedIn and X. Always happy to talk agents, agentic architectures, or how AI can fit into a real business workflow. My DMs are open.