Hi, I'm Sanjay. I'm a 3rd-year CSE student at Amity University Rajasthan. I like building AI systems that do real work, and the backends that keep them honest.
- Agents with limits. Retry caps, loop guards, compliance gates, and a human in the loop for risky calls.
- Reliable tool use. Making MCP and LangChain tools behave when an LLM is the one calling them.
- Grounded answers. Citations, source checks, and documents that replace each other over time.
- Visible reasoning. Full traces, so I can see why an agent did what it did.
- FastAPI backends that carry state, retries and an audit trail.
- Agent workflows for finance and compliance, where mistakes cost money.
- Small, readable code I can explain line by line.
- Making agent and tool workflows dependable enough to leave running.
- Turning "it worked once" into something I can test and trace.
Live demo · Source · runs on your own Gemini key
- Build it, then try to break it. Weak spots show up faster that way.
- An agent without limits is a liability. Cap it, gate it, and let a human decide the risky calls.
- If I can't trace it, I don't trust it.
Languages Python C++
Backend FastAPI
AI Machine Learning Deep Learning Generative AI Agentic AI RAG
Tooling LangGraph LangChain FAISS
Contributed to AMD GAIA, AgentAcct, OPCUA-MCP and ResultSeal. Reliability fixes, agent tooling, MCP tooling and LangChain integrations.
Got a flaky agent? Send it over.