AI Products · Final-year CS · Nextleap PM Fellow

Start with the user.
Add AI only where it helps.

I'm Mrigank, and I build AI products. I figure out what an AI feature should do, where AI shouldn't be used at all, and how to measure whether it's actually helping users.

See my work ↗ Let's connect
Selected work

What I built, and what I didn't

Six products, AI first. For each one: the obvious version I didn't build, what I built instead, and the number that tells me it's working.

LiveFrom a briefRAGEvals

HDFC Fund Facts

A RAG assistant for five HDFC mutual funds that cites a source for every answer and refuses anything that sounds like advice.

The obvious version

A finance chatbot that compares funds and calculates returns

A calculated return can't be traced to a source document, so nobody can check it. It answers in three sentences or fewer, links exactly one source, and declines anything that sounds like advice.

Also left outReturn calculationsFund comparisons

It's working if target

0

answers on the test set say something the source pages don't.

While catching over 95% of advice-seeking questions.

Read case study↗ Try it
LiveLLMVisionHealth

BeyondLabel

An LLM that reads a food label and tells you what it does to your body, and what to eat instead.

The obvious version

An AI that simplifies the nutrition label

People weren't misreading labels. They thought muesli was healthy, or knew chips were bad without knowing how bad. So it explains the effect on your body, and suggests a swap for the same craving.

Also left outLive barcode scannerWhatsApp botCalorie logging

It's working if target

1 in 4

scans end with the person choosing the suggested swap.

Without letting a verdict take longer than 10 seconds.

Read case study↗ Try it
From a briefAI agentMCP

App Review Pulse

An AI agent that turns hundreds of app-store reviews into a one-page weekly note, delivered through MCP.

The obvious version

A review dashboard with sentiment charts

Nobody opens a dashboard on a busy week. A 250-word note with the top three themes, three real quotes and three actions lands where the team already reads: Google Docs and Gmail.

Also left outMore than five themesReviewer names

It's working if target

1+

action from each weekly note makes it onto the backlog.

With no personal data in any note.

Read case study↗
LiveExperimentationML

A/B Decision Engine

An experimentation engine with an uplift model that tells a team whether to ship, kill or keep testing, and whether to trust the result.

The obvious version

A calculator that checks if p is below 0.05

Real tests go wrong before significance matters: a broken traffic split, a team peeking daily, an average that hides who got hurt. It checks all three, then gives one verdict.

Also left outA dashboard before the maths was proven

What I measured measured

22.7% → 1%

false-positive rate from daily peeking, before and after sequential testing.

Stays under the 5% error rate it promises.

Read case study↗ Try it
LiveFintechNo AI, on purpose

Plutus

A money app where we deliberately left AI out, so every number is arithmetic you can check.

Team project with Kaustubh Gaurav, product designer. I owned product.

The obvious version

A budgeting app with bank sync and an AI advisor

The problem was losing track of small daily spends and loans between friends. So logging takes seconds, every number is plain arithmetic, and money owed to you never looks like money you owe.

Also left outBank syncAI advisorForecastingAccounts

It's working if target

5 of 7

days in a new user's first week have at least one expense logged.

With no number on screen that can't be traced to a saved record.

Read case study↗ Try it
LiveB2B SaaS

Kudos

Peer-to-peer recognition and rewards for teams.

The obvious version

An HR console where managers hand out rewards

When only managers can give recognition, they become the bottleneck and the people who see the work stay silent. So anyone can recognise anyone, on a ledger that can't lose a transfer halfway.

Also left outHR system integrationsReal gift-card payoutsNotifications

It's working if target

40%

of a team gives kudos every week by week six.

With zero points-balance mismatches.

Read case study↗ Try it
Case studies

Problems I dug into,
and what I'd build

From user research to tested prototypes: the problem, the fix, what I rejected, and the number that decides.

Tested prototypeNextleap

Google Photos: people don't search old photos. They scroll.

Built: a timeline that knows people. Tap a face to see everywhere they appear.

TestedWrong photos opened fell from 8 to 0 across 10 testers.

Read the case study
Live prototypeNextleap

Myntra: they want it. They're unsure about the price.

Built: a WhatsApp price-drop alert backed by the item's real price history.

Decides itPurchase rate after a price verdict, tracked per zone.

Read the case study
AI productInsurance

Cashless claims: the paperwork's done, the family still waits

Proposed: an AI layer that builds the claim from admission, ready before discharge.

Decides itSettled within 10 minutes of discharge, approval rates unchanged.

Read the case study
User researchSurvey, n=34

ChatGPT Voice in India: 97% knew it existed. 71% still typed.

Proposed: suggest voice where it clearly wins, and let people edit before sending.

Decides itVoice sessions per daily user, total messages held.

Read the case study
TeardownProposal

MyJio: your data balance, without the ads in the way

Proposed: a home-screen widget built on data the app already caches.

Decides itOver 5% adoption in 14 days, ad views down under 8%.

Read the case study
TeardownProposal

Zomato: hungry, scrolling, still can't decide

Proposed: "Help me decide": three picks, each with a reason.

Decides itBrowse-to-order conversion, order value held.

Read the case study
Tools I think with

The toolkit behind
the thinking

Every product decision needs the right tools. These are the ones I use to research, wireframe, analyze, and ship.

See how I use them ↗
Tools I think with

The toolkit behind
the thinking

Every product decision needs the right tools. These are the ones I use to research, wireframe, analyze, and ship.

Figma
Notion
Excel
Canva
Google Analytics
Jira
VS Code
Claude
Supabase
PostHog
Sentry
Apify
Vercel
See how I use them ↗
MRC
The person behind the product

A CS student who learned product by shipping AI,
with a finance habit on the side.

I learn fastest by building something small, putting it in front of people, and watching what they actually do with it. I'd rather spend a week understanding a messy problem than a day building the wrong thing.

Final-year CS at KIIT, with an FMVA and six months at a CA firm on the side.

MRIGANK RAJ CHOUHAN
AI Product Builder · Final-year CS
Institution KIIT Bhubaneswar
Degree B.Tech CSE (Final Year)
Academic CGPA 8.36 / 10
Fellowship Nextleap PM Fellow
Finance Cert. FMVA — CFI
Graduating 2027
Let's talk product ↗
Experience

Where I've
learned by doing

Product Management Fellow
Nextleap
Product discovery, PRDs, metric design and user research, taught by product managers from Indian startups.
2026 – Present
Fellowship
Finance Intern
Rahul Singh & Associates · CA Firm
Reconciled invoices and prepared compliance documents for small-business clients. Saw first-hand the manual workflows fintech products exist to replace.
Jun 2025 – Dec 2025
Finance · Ops
Equity Research
Independent · HDFC, Reliance Industries
Built a DCF model with sensitivity tables for Reliance Industries, and wrote a research note on HDFC Bank.
Jan 2026
Finance · Research
B.Tech Computer Science Engineering
KIIT Bhubaneswar · CGPA 8.36
Final year. Coursework in data structures, machine learning and software engineering. Graduating 2027.
Jul 2023 – 2027
Education
FMVA Certification
Corporate Finance Institute (CFI)
Financial Modeling & Valuation Analyst: DCF, comparable company analysis, LBOs and financial statement analysis.
Jun 2025 – Feb 2026
Certification
Off the clock

Things I'm still
figuring out

Product, tech, and the occasional detour into life. I write to think clearly, and publish it on Medium.

Read on Medium↗
Get in touch

Tell me what
you're building.

Hiring for a product role, or working on something with AI? I'd love to hear about it.

Write to me
mrigankrajchouhan@gmail.com
Elsewhere
LinkedIn ↗ GitHub ↗ Resume ↗

Or send a note.