Products
that move
with users.
Product-minded builder who ships AI products end to end — user research, sharp prioritisation, and the code to prove it.
Featured case study
Trust,
by design.
How I designed an autonomous AI voice agent that merchants can actually trust with their money — by deciding exactly where the AI is not allowed to decide.
Read the case studySelected work
Visit live site ↗PayRevive
AI Hinglish voice agent for Razorpay merchants that detects failed payments, diagnoses the cause and calls customers to recover them.
The callLet the AI act on its own only inside a deterministic policy engine — a ₹50K ceiling plus a full audit trail — so merchants can trust it.
Visit live site ↗PrepSense
AI resume analyzer and voice interview coach: ATS checks, bullet rewrites and voice mock interviews with progress tracking.
The callMade it role-agnostic instead of tech-only, so any job seeker gets field-specific feedback and questions.
Real Estate Platform
Property listing and lead-management web app for a real estate business, with searchable browsing and lead capture.
The callBuilt v1 around the two things that drive the business: finding a property fast and capturing the lead.
Illustrative mockup · client under NDAUniform Shop Inventory
Inventory management system for a school uniform shop, with barcode scanning to track stock by item and size in real time.
The callTracked stock per item and size, with barcode scans at the counter — so the owner knows which sizes to reorder before the back-to-school rush.
Case study
PayRevive · Razorpay AI Buildathon · Track 3: AI Revenue Recovery01 / 04
The problem
Failed payments are silent revenue loss for merchants. Recovering them means noticing the failure, understanding why it happened, and reaching the customer quickly — work most small merchants don't have time to do.
02 / 04
Key decisions
Every design choice came back to one question: would a merchant let this agent call their customers?
03 / 04
How it works
Payment events come in from Razorpay, get verified, classified and routed through the policy engine before any call is made.
- HMAC-SHA256 signed webhook verification — only genuine events trigger action
- Gemini-based intent classification of the customer's response
- Policy engine decides what the agent may do autonomously
- Every step written to an audit log for explainability and compliance
04 / 04
What I'd measure next
To take it from buildathon to production, these are the metrics I'd track first.
- Recovery rate — % of failed payments recovered after a call
- Call completion & drop-off point in the conversation
- Escalations — how often the agent hands off to a human
- Merchant trust — share of merchants who keep autonomy switched on
Experience
- Feb 2026 — NowAI TraineePW (PhysicsWallah) · Apprenticeship · Noida, Hybrid
- Contributed to 20+ AI filmmaking projects for the PW App / PI section, supporting AI-enabled content production
- Researched and tested 30+ AI tools to inform content-production workflow decisions
- Coordinated with 8+ creative and production team members; managed timelines, deliverables and stakeholder communication
- Tracked content metrics and contributed to improving content performance
- Jan 2025 — Jan 2026Associate Product ManagerHighFurries · Dwarka, Delhi
- Ran user research and competitor analysis to define requirements and prioritise the roadmap using RICE and MoSCoW for a 2–3 person product team
- Owned research for the doctors database and partnered with design and engineering on the “AI Disease Scan” feature from requirements through deployment
- Supported third-party API integrations and frontend performance optimisation as part of feature delivery
- Planned and ran “Paint with Pets”, a live customer-engagement event that achieved an 80% conversion rate
- Apr 2025Winner, HackQuest HackathonMSIT
- Built BlockEd — gamified learning with NFT on-chain certificates (React, TypeScript, Solidity)
- 2022 — 2026B.Tech, Computer ScienceNetaji Subhas University of Technology, Delhi
Product teardowns
10 decks · real companies, public data- 01Problem
- 02Users
- 03Journey
- 04Pain points
- 05RICE
- 06MVP
- 07A/B test
- 08Go-to-market
- 09Risks
PhysicsWallah
Why are daily users up 26% while paid users are up only 2.4%?
+26% vs +2.4%daily vs paid users, Q1 FY27
- Test first
- Batch Finder — a short quiz and diagnostic that recommends one batch on one card
- A win looks like
- +10% payment conversion, refunds and retention flat
Zepto
Why does India's second-largest quick-commerce app have the smallest basket?
₹387average order, vs ₹665–700 for rivals
- Test first
- Smart cart rail of cheap, relevant add-ons, plus an all-in price
- A win looks like
- +₹20 AOV with delivery time and margin flat
Zomato
Fees keep rising. How do you keep people ordering as often?
₹2 → ₹14.90platform fee in three years
- Test first
- All-in price shown from the menu, not only at checkout
- A win looks like
- +2 pts checkout conversion
Uber
When a driver cancels, how do you keep the rider from switching apps?
₹1,511 crUber India FY25 loss
- Test first
- Fare lock on re-match after a driver cancels
- A win looks like
- +2 pts 7-day rebooking, driver earnings flat
Airbnb
Airbnb booked the stay. Who books the rest of the trip?
~35%of first-time hotel guests later book a home
- Test first
- A “Plan your trip” card with three nearby services after booking
- A win looks like
- More extras per 100 stays, cancellations flat
PlayStation
When prices rise and playtime dips, how do you keep PS Plus members renewing?
−36%PS5 shipments in the quarter
- Test first
- A value recap shown before renewal
- A win looks like
- +2 pts renewal rate, revenue per member flat
GitHub
When AI makes pull requests free to write, who pays the cost of reviewing them?
518.7Mpull requests merged in a year, up 29%
- Test first
- Opt-in PR readiness gate with a what's-missing checklist
- A win looks like
- Fewer review hours per merged PR, first-timer merges down ≤5%
How does WhatsApp keep the inbox trusted while business messaging grows?
99.67LIndian accounts banned in January 2025
- Test first
- Default-on message requests for unknown numbers in India
- A win looks like
- Fewer scam reports per 1,000 users, business read rate flat
How does Instagram keep friends visible as Reels takes over?
51%of US Instagram time now spent on Reels
- Test first
- An “Open Instagram to” setting: Home, Following or Favorites
- A win looks like
- More friends-first sessions, time spent and ads flat
Snapchat
How does Snapchat keep friend groups active where each user is worth the most?
−7%North America daily users, to 92M
- Test first
- A weekly reconnect nudge naming an active close friend
- A win looks like
- More lapsing users back weekly, opt-outs and uninstalls flat
Toolkit
PM craft, backed by dataProduct
Research & Analytics
Execution
Technical & Tools
Get in touch