Sole PM from pre-launch through exit on Kotak Bank's premium-card travel rewards platform. Defined the platform construct end-to-end: partner API integration layer for flight and hotel suppliers, event-driven state management across booking and payment flows, lifecycle notification stack. Flights first, then hotels, then points-to-airline-miles redemption. Scaled from zero to ~₹85 Cr annualized GMV run-rate — 220x monthly growth in 12 months with ~12,000 premium cardholders transacting.

The ops team was spending 7 hours daily sifting through 50–90K customer events manually. Built a four-agent system using CrewAI, Groq inference, and BERT-based sentiment classification — topic modeling, regression detection against release notes, insight generation. Deployed on AWS, used org-wide. Discovery time down to 40 minutes. CSAT up 40%.

Identified a D2C distribution opportunity: become our own gift card issuer rather than a reseller procuring from QwikCilver. Owned the full 0-1 build — RBI compliance constraints defined the architecture (KYC tiers, PPI categories, transaction limits, reconciliation flows). Shipped in 3 months with 2 engineers and 3 QA. Partnered with 200+ D2C brands. Generating INR 40 crore in year-one revenue.

Super-premium users on HDFC Smart Buy were consistently searching for products outside the catalog — high-value items like Omega Seamasters, premium appliances, limited inventory. Built a reverse marketplace where every step has an intelligence layer: search intent classification, feasibility scoring before showing the request form, vendor selection by capability match and reliability, bid ranking by margin and fulfillment timeline, outcome-driven feedback loops. Not a form — a chain of scoring decisions. Runner-up at the R360 AI Hackathon, subsequently funded for productionization. INR 8 crore projected first year.

Owned a flights product serving 100K+ users across Reward360's rewards platforms. Improved end-to-end conversion by 12% through funnel analysis — payment failures and unclear cancellation policies were the biggest drop-off points. Built a white-label version so client banks could deploy under their own brand, shifting our margin from 3% (affiliate) to 6% (platform owner). Contributed to 15% quarterly booking revenue growth.

Agents were approving refunds by default because they had no objective information to decide otherwise. Built a six-layer deterministic risk scoring and decision support system: experience-level anomaly check, policy gate, customer risk profile, request evaluation, classification, and live case management — with a Claude-powered LLM layer for agent communication and evidence synthesis. ~88% reduction in agent hours; fraudulent approval rate from ~5% to ~2% target. System cost: $78/month.

The loyalty team was spending 80% of their time on manual reconciliation — thousands of records matched by hand across systems. Identified the bottleneck, scoped the automation, shipped it. Processing time from 8 hours to 15 minutes. INR 3.4 crore saved annually. Freed the team to do the work that actually needed humans.

I own payment experience across two platforms: Kotak Unbox and the multi-partner flights vertical. On Kotak, I maintain in-app checkout (users stay inside the bank app rather than getting redirected to a generic gateway), saved-card transactions for repeat bookings, and flexible fare selection to reduce transaction friction. On flights, I run multi-partner journey orchestration across ClearTrip, EaseMyTrip, and other affiliates for the same listing — plus failure handling that surfaces clear error states, retry logic, and messaging that tells users what went wrong and what to do next. Payment failure was the single largest drop-off point in the flights funnel; the orchestration work is what drove the majority of the 12% conversion improvement.