Rishita Mantri

Data, analytics and AI. I moved from India to the US, studied Computer Engineering at NMIMS and Quantitative Management at Duke, and now build AI agents. I can't stop asking why.

How I got here

Adaptability

I moved from India to the US with a lot of questions. New country, new friends, new way of working. Turns out I like figuring things out from scratch.

Caring about people

In my second year of college, I taught kids who didn't have much access to good schooling. They asked better questions than most adults. I still think about them when I explain anything.

Curiosity

I can't stop asking why. Why do people comment the way they do? Why does one product feel easy and another doesn't? Data is just my excuse to keep asking.

Projects

Adaptive Financial Analytics Agent

An AI agent that turns any financial dataset into the right KPIs and a ready dashboard.

Before, someone had to study each financial file and decide what to measure. Now the agent does it for 8+ types of financial data, building a full dashboard for under 10 cents per file, with every number traced back to its source.

Tools I used: Cursor, Claude

The problem

Every financial file is different. A profit and loss statement needs different metrics than an invoice list or a bank statement, and teams spend hours deciding what to track before they even start analysing.

What I built

An agent that reads the file, figures out what kind of financial data it is, picks the KPIs that matter for it, and builds the dashboard. It asks a clarifying question when something is unclear, like whether invoices are inbound or outbound. Underneath, an engine calculates and validates every number with code and traces it back to its source cell, so nothing is guessed. It handles 8+ types of financial data today.

What I learned

AI is great at deciding, but numbers should be calculated, not guessed. Splitting the two made the results accurate and consistent every time.

Yankees fan sentiment and segmentation

Understanding who Yankees fans are and how they show up across platforms.

The team could see how much fans engaged, but not why. I turned 30,000+ messy social media comments into six fan segments, each with its own strategy, and presented them to senior stakeholders.

Tools I used: Python

The problem

Engagement numbers showed how much fans interacted, but not why, or how they actually felt about the brand.

What I built

I analysed 30,000+ posts and comments across Instagram, Reddit and Twitter, and built a sentiment model tuned for how fans really talk online, slang and sarcasm included. From that, I identified six fan segments and shaped a tailored engagement strategy for each, which I presented to senior stakeholders.

What I learned

People rarely say what they mean literally. "This game was sick" is a compliment, and a model has to learn that before the numbers mean anything.

Global Portfolio Revenue Analysis

Executive dashboards for a 66-store global portfolio.

Leadership saw one big revenue number. I showed them that three categories drove 67% of it, and flagged 17 stores quietly putting 21% of revenue at risk.

Tools I used: Excel, Tableau

The problem

Leadership needed to know where to invest and which stores were quietly struggling.

What I built

Tableau dashboards analysing $112.3M in gross profit across 66 stores. They showed that the top 3 categories drove 67% of revenue, and flagged 17 at-risk stores that made up about 21% of six-month revenue.

What I learned

A dashboard is only useful if the person reading it knows what to do next, so I designed it around decisions, not charts.

From my diary, November 2026

Things are moving so fast right now. People who have never written a line of code are launching websites in an afternoon. Honestly, it feels a little surreal. I spent years in computer engineering learning how to do exactly that, and now anyone can.

At first it felt like a curse. If everyone can build, what's left for me?

But I think the blessing is hiding in plain sight. When everyone is fast, speed stops being special. What matters is judgment. That means knowing what to build, and also looking at the tools you already have and figuring out how to make the most of them for the business.

I was working on a SQL agent that anyone, anywhere could use, and I realized something while testing it. At one point, the AI wrote a query that could have deleted the entire dataset. It didn't run, because I had built a validation layer that checked every query before it touched the data.

I didn't add that layer because I saw this coming. It was just a habit from engineering: never trust something without checking it first.

That's what stays with me. Someone playing around with AI on their company's data might not have that habit. One confident query, and everything is gone.

So maybe those years weren't just about learning to code after all. AI can do the typing now. What I really learned was how to think, and that part still matters.

Contact

  • Email: rishita.mantri@duke.edu
  • LinkedIn
  • GitHub
  • Resume (PDF)
? ?

One sec, finding my keys...

Resume

    Tap or scroll to open the house