Headline

Software Engineer @ Eli Lilly | GenAI • Agentic AI • RAGs • AWS • LangChain • LangGraph • OpenAI SDK• Immuta

About

Software Engineer at Eli Lilly building enterprise GenAI systems, AI governance workflows, and cloud-native integrations on AWS. My work focuses on designing secure AI-enabled enterprise platforms using Immuta, AWS, and modern LLM frameworks. I work extensively on AI onboarding workflows, data access governance, API integrations, and scalable backend systems for enterprise data platforms. Current areas of work include: • Agentic AI and MCP-based workflows • Building custom MCP servers for agentic AI workflows, tool calling, enterprise integrations, and contextual AI orchestration • AI-powered onboarding and governance automation • Immuta integrations and webhook-based attribute management • S3 Access Grants and secure enterprise data access workflows • RDS consumption APIs and enterprise backend integrations • Custom UI and access provisioning systems • AI coding assistants and automated PR review systems • Enterprise GenAI applications using LangChain, and custom MCP servers Tech Stack: AWS Lambda • API Gateway • S3 • RDS • IAM • Python • FastAPI • LangChain • LangGraph • Immuta • MCP • RAG • REST APIs • Splunk • Immuta Interested in building scalable, secure, and production-grade enterprise AI systems.

Location

Bengaluru

AI stack

anthropic apis

claude code

hugging face

langgraph

opencode

claudecode

Work experience

AI Engineer at Eli Lilly and Company, Aug 2024 - Present · 2 y 2 mo, Bengaluru, Karnataka, India, I joined Lilly in August 2024 as a software engineer on the AI side, and over the last two years I've built three production systems. Each one replaced a slow, manual process with an agent that has guardrails. The first was Autonomous Operations. IT requests were taking three days or more, so I architected a multi-agent pipeline. Rules handle the clear cases, embeddings handle the next layer, and Claude is only the last-resort tiebreaker. Every request goes through SOP prevalidation and human approval before it executes, and every decision is stored with its reasoning. Turnaround dropped to 1 to 2 hours. The second was a data discovery agent. Finding which S3 datasets you could access used to take weeks or months of chasing people. I built an MCP-based agent on top of S3 Access Grants and Immuta mappings, deployed it on Kubernetes with CI/CD, and discovery became near-instant. Nobody handles raw credentials either. The third was a self-service portal for onboarding data-access use cases. The process was scattered across Immuta, ServiceNow, IDM, and Azure, and took eight weeks. I built a guided portal that consolidated it and auto-generated the policy summaries, test matrices, and Jira tickets. It now takes under a week. The common thread is that I pick a painful process, design the architecture, ship it to production, and keep a human in the loop where it matters. That's the work I want to do more of, faster, at a startup.

Software Development Intern at Aithent Inc., Mar 2024 - Jun 2024 · 3 mo, Gurugram, Haryana, India, I did an internship at Aithent, a few months in 2024. They had developers at client companies digging through product manuals and PDFs to find answers, so I built a chatbot they could just ask in plain English. I did the whole retrieval side myself. I pulled in the documentation, chunked it up, stored the embeddings in a vector database, and used similarity search to find the most relevant passages. Those went to GPT-3, so it answered from the docs instead of guessing. It was my first real project with LLMs. Honestly, it's what got me hooked on building AI that people actually use.

AI Developer & Trainer (Freelancer) at Remotasks, Remote

Education

Bachelor of Technology - BTech, Data Science Engineering, Manipal institute of technology

High School, CBSE, Mordern Vidya Niketan

Public links

github: https://github.com/yashsinha1611