I build the systems that make growth work, then scale spend on top of them. A creative-scoring model that filters bad images before the auction. CAPI wired to CRM so bids chase qualified leads. An incrementality framework that found ₹10L/month of spend the business didn't need. Today that runs ₹3.5 Cr/month at Housing.com — before it, I co-founded Flex EV and built its fleet platform from scratch.
Headlines first. Open any one of them for the cases, the numbers and how it was actually done.
Built the measurement and creative systems first — then scaled spend on top of them. 2× SPOTLIGHT Quarter Award.
Budget allocation running on last-click and gut calls.
Built the cross-city testing framework Housing.com now uses to measure true campaign lift and cost per incremental lead, across 12 markets. A geo-holdout in Chennai surfaced ₹10L/month of non-incremental spend — roughly ₹80L/quarter at full scale.
Bad property images slipping into live campaigns, week after week.
Shipped 15,000 videos through Kling AI and Veo 3 and 30,000 enhanced images. Built a runtime ML scoring model with the Data Science team that strips non-marketable creative from the feed before it serves, and expanded into Reels and Stories.
Campaigns optimising on broken signal — most of the catalog never matched.
Stripped irrelevant Meta Pixel and FB SDK parameters, added an hourly out-of-stock cron, and enriched property params. Fixed an 18% city-mismatch rate in the same pass.
No per-channel × campaign view of what a good lead actually looked like.
Connected CAPI for Leadgen to CRM and dialer signals, then moved optimisation onto the call-connected lead rather than the raw form submit.
Callers working manual lists, with no view of agent productivity.
Built the CRM and the automation layer around it for a 5-person lead-generation calling team — the work that won the first SPOTLIGHT award.
A fragmented 180-campaign account starved of learning volume.
Restructured 180 → 100 campaigns into PAN-India, city-cluster and city tiers, and shifted bidding from Highest Volume to tCPA under SOP-led governance. CPL −5% QoQ, CPM flat YoY against an industry that rose ~5%, affiliate share 6% → 9%.
Co-founded an EV rental business for gig delivery riders, and ran it alongside a full-time role. Built the ops, the platform and the P&L.
Co-founded a B2B2C EV rental business for gig delivery riders in Delhi NCR. Scaled to 500+ vehicles and ₹40L MRR in 12 months at 86% utilisation — while holding a full-time role at Housing.com. Stepped back in October 2025.
Rider onboarding, collections, recovery, maintenance. 1 → 3 hubs, 2 → 30 team.
Replaced manual QR collections with Easebuzz bulk payment links and automated rider KYC through Leegality. Cleaned up reconciliation, cut leakage.
Started the whole operation on Google Sheets, GForms and Apps Script, then rebuilt it as an in-house Fleet Management Platform and Rider App for rent tracking, dues and asset visibility.
Piped third-party in-vehicle IoT and webhooks into our own platform for remote immobilisation and theft prevention.
Cold outreach → commercial close across three delivery platforms, plus OEM and battery-swap supply deals.
Hired the first hub managers. Wrote the SOPs for collections, recovery, redeployment.
Stepped back in October 2025 after a structural disagreement on operating discipline with my co-founder — foundations before scale. Took the lessons; took the operator scars.
Their first-ever PM, out of IIT KGP. Took reporting from on-prem Excel to a production data platform on AWS — and co-authored the AWS blog about it.
My first role out of IIT KGP — and Pinnacle's first-ever PM, in their R&D wing. A BIM engineering firm running ~500 live projects for international clients. I helped take its reporting from on-prem Excel to a production data platform on AWS.
Dashboards running straight off live production databases — slow loads, constant engineering drag.
Shipped a cloud BI stack on AWS S3 + Glue + Athena + QuickSight and migrated 6+ dashboards off on-prem, then specified a natural-language chatbot that turns plain-English questions into charts on the same warehouse. Gave engineering back 6 man-hours a week.
Co-authored the AWS blog on this work ↗
Project staffing made on tribal knowledge across a global engineering org.
Built a feature store and shipped a Talent Tracking System to match engineers to projects, surface hiring gaps and plan training. Resource allocation went from 3–4 days to hours.
Client Desk had low adoption among 400 named accounts.
Led the activation campaign, reworked onboarding. DAU grew 70%+ QoQ; onboarded 98 of 400 clients in a single month.
A product scoped end to end — market, segmentation, revenue architecture, unit economics, roadmap. Modelled, not shipped.
Indian brokers fight over a shrinking pool of portal leads, then struggle to find the right inventory once a buyer is finally in hand. Solving demand alone is a feature. Solving supply alone is a feature. BrokerBoost resolves both sides of the same transaction — a broker runs their own campaigns from one console, and when a buyer turns up, a co-broking handshake surfaces matching listings across the network in seconds.
Blasts to the broker's own lead lists at ₹2/msg, top-up credits, no lock-in. Telco pass-through pricing, ~50% margin.
Auto-creative and targeting at ₹240 / 1K impressions. Media passes through; the margin sits in the layer above it.
Auto-generated listing brochures, reels and broker brand kits at ₹199/pack — margin expands as model costs fall.
A ₹200-a-side handshake matching one broker's buyer to another's inventory. Pure software, bilateral lock-in, ~95% gross margin.
Premium brokers already buy portal packages; volume brokers spend ~₹1K/month. The mid tier is the wedge — digital enough to adopt, under-served enough to care.
Modelled revenue per lead channel by channel against the portal's own — the deck answers "doesn't this cannibalise us?" with arithmetic rather than assertion.
This is a concept deck, not a shipped product — the market sizes, pricing and three-year trajectory in it are modelled, not measured. What it demonstrates is the reasoning: segmentation, wedge selection, revenue architecture, unit economics and an engineering roadmap, argued end to end.
Co-authored with AWS for their Business Intelligence blog — how we took Pinnacle from on-prem reporting to a QuickSight data platform. Published Dec 2023.
Read on aws.amazon.com↗A structured product case on a real Flipkart problem — retention, growth, user-journey diagnosis, and a recommendation that lands.
View deck↗Attribution tells you which pipe the water came out of. Incrementality tells you if you needed the pipe. Once you've run a geo-holdout, watched a channel go dark, and found ₹10L a month the business never missed, you stop arguing about last-click.
Most "AI in marketing" posts are prompt libraries. The real wins are boring: scoring creative before it ships, generating 15K variants when you'd previously brief 15. Hype is on image generation; leverage is on the filter that runs after.
Most growth roles hand you a machine and ask you to turn the dial. The interesting work is upstream of the dial — the feed, the signal, the measurement. Fix those and the dial starts working; skip them and you're optimising noise. I'd rather own the thing end to end than tune someone else's.
Currently at Housing.com, three years in. Open to growth / performance marketing and product roles at mid to mid-senior level. IIT Kharagpur '23. Gurugram, open to relocation.