SAI

Sai

Computer Science & Economics — building AI systems and machine learning models.

I build things that turn ideas into solutions.

01

What I build

I got into software because I like taking something broken or unnecessarily complicated and making it work the way it should, for people who don't have the time or patience to fight through it themselves. I like building sustainable solutions, not quick fixes.

A key project of mine was SignBridge, a real-time ASL fingerspelling recognizer built with MediaPipe hand tracking and a Random Forest classifier, streamed over WebSocket and served through Flask. A five-frame majority vote smooths out noisy predictions, which gets it to 94.2% live accuracy. More recently I built a Multi-Agent Research Analyst, a LangGraph pipeline where a planner, retriever, and verifier hand work off to one another, with LangSmith tracing and a ragas evaluation suite so I can actually check what the agents are doing instead of trusting the output blindly.

That same instinct (take something complicated and make it legible) pulled me toward the business side of building too. I'm a Venture Lab Fellow at Grey Matter Society, developing BillFight, an AI powered platform that helps people dispute medical bills, accepted into Rutgers' Launch Circuit accelerator.

This year I joined MSH Global as an AI Engineer Intern, building a retrieval augmented generation pipeline so people can ask plain-language questions about dense insurance documents and get grounded answers back.

School
Rutgers University
Studying
CS, minor in Economics
Graduating
2027
Standing
Dean's List
Focus areas
Applied AI, software engineering, machine learning, product
02

Experience

A mix of internships, fellowships, and things I built on my own.

Building a retrieval augmented generation pipeline for insurance document Q&A, so people can ask plain-language questions about dense policy documents and get grounded answers back.
Developing BillFight, an AI powered platform for disputing medical bills, accepted into Rutgers' Launch Circuit accelerator.
Worked across the stack in Java and SQL, writing JUnit tests and shipping on an Agile team.
Real-time ASL fingerspelling recognition using MediaPipe hand tracking and a Random Forest classifier, holding 94.2% live accuracy with five-frame majority voting.
A LangGraph pipeline where a planner, retriever, and verifier hand work off to one another, with full LangSmith tracing and a ragas evaluation suite so agent outputs can be checked, not just trusted.
Behind the scenes
how the multi-agent pipelines I build actually think
Plan Retrieve Verify Respond
03

Let's talk

I'm a student who likes building things that make complicated systems easier to deal with. Always happy to connect.