01About
I build AI systems for problems that already have victims, and I build them so a sceptical reader can check every claim.
In brief
I am a final-year computer science student at The LNM Institute of Information Technology, Jaipur, graduating in 2027. Over the last two years I have shipped 32 public products and competed in more than twenty hackathons and datathons, and almost all of it points at the same kind of problem: something in the physical world that is already going wrong for somebody, where a model can help but cannot be trusted on its own.
Crop loss. Air quality enforcement. Road capacity. Fraud rings. Misdiagnosis. Thin-file credit. Urban flooding. Police case linkage. The common thread is not the technique, it is the discipline: a model is allowed to propose, and something deterministic decides, so a person can always answer the question “why did this happen?”
Outside the engineering I spent three years in the Debate Society, ending as one of its Coordinators, and helped run two editions of the LNMIIT Model United Nations. Running a conference for 170 delegates taught me more about shipping under a deadline than any side project did.
shipped products
hackathons & datathons
packages on PyPI
best Orchestrate rank
Experience
Full Stack Engineer, Agentic AI (Intern)
Compared three ways to run LangGraph agents in production and chose Aegra, the open-source Agent Protocol server, because the commercial platform would move agent state outside Deloitte and an in-house runtime would rebuild a mature tool.
Built capability in five layers on a multi-agent travel planner, from a single agent to a supervisor with three specialists, a human approval gate, real browser tools over MCP and serving through Aegra.
Replaced request-time graph assembly with a graph factory that registers one static graph and carries per-request variation in run state, verifying the migrated output identical to the original field by field.
Kept the public API unchanged so the front end and test collections needed no change, and made the migration reversible through a single setting, with the original execution path kept byte for byte.
Proved it on the live SAP tenant: a Bank Accounting goal became 8 configuration tasks, run in a real browser through Playwright MCP, recovering from a duplicate entry and a mis-filled field to finish 8 of 8 in about 15 minutes.
Worked out a Docker-free deployment on a shared Linux server and fixed failures that raised no error, from two agent versions colliding on one module name to a stream reading a field the SDK never returns.
“I came in knowing how to build agents. I left knowing how to make them survivable.”
Education
B.Tech, Computer Science and Engineering · Jaipur, Rajasthan
Final year. The sixth-semester B.Tech project became AGENTIQ, an agentic platform for autonomous API testing, security validation and deployment assistance. Outside the classroom, most of my time went to the Debate Society.
Intermediate, Science with Mathematics
Physics, Chemistry and Mathematics.
ICSE, Council for the Indian School Certificate Examinations · Gorakhpur, Uttar Pradesh
The top grade in Mathematics, Computer Applications, Hindi, and History, Civics and Geography, with grade 2 in English and Science.
Outside the terminal
public repositories
Reading, building and breaking things. Most are deployed and free to use, with the failures published beside the results, and three are packages on PyPI.
years in the Debate Society
From first-year member to Coordinator, and seven Model United Nations conferences on both sides of the table.
committees run in one weekend
The reading behind the debating: the South China Sea, delimitation, territorial sovereignty. It is also why most of what I build points at a public problem.
years with Imagination
The campus photography club. Mostly events and people, and some cinematography on the side.
Competing on real data is the quickest honest feedback there is. Currently working through Kaggriculture.
The one thing on this page that has nothing to do with a screen.