Stacked benefits experience
One modernized platform behind a new benefits experience, with smart spend-down that weighs your accounts, expiring funds, and each expense so you don't have to.
Product Manager
Hi, I'm Rohan Pandit
Product strategist and engineer at heart, building the bridge between complex systems and the people who rely on them.
I love to build. I started as an application developer before moving into product, an engineering foundation that still shapes how I think about systems, not just a roadmap. In 8+ years at Optum Financial, Starship, and Engine Room, I've led product builds across fintech, healthcare, and ecommerce for millions of consumers.
Most recently, I've applied that same drive to AI: shipping agentic systems with tool-calling, multi-agent orchestration, RAG pipelines, and guardrails/evals, because I wanted to prove I could build it, not just talk about it.
The problems I find most exciting are the ones without a playbook yet — new technology, real consequences, and not a clear consensus on the right move.
Specialties
Selected work
Case Studies
Shop
A new online shop that helps members maximize their supplemental benefits, like OTC and healthy food, on eligible products fulfilled by Amazon or Walmart.
Wanderwise
A travel-planning concierge that helps you plan your dream trip.
Stacked benefits experience
3new clients
$1.5MAUM growth
call volume
stuck claims
stuck claims
Benefits are confusing. People didn't know how their accounts worked, what was covered, or how to handle a claim, so money went unspent, claims stalled, and members picked up the phone for help. That friction kept utilization low and put contract renewals at risk.
Using user research and call data, I wrote the product brief covering strategy, data flow, and architecture to modernize the experience and consolidate separate platforms into one. The new experience does the hard part for members: smart spend-down logic weighs which accounts they have, when funds expire, and what each expense is, so their money gets used the right way. Consolidating also opened a new distribution channel.
- Grounded in the data: ~42% of members left money unspent at year-end (~$380 on average), ~15% of claims got stuck in substantiation, and ~25% led to a support call.
- The new distribution channel brought in 3 new clients, $1.5M in AUM growth, and 6,000 new users.
- Fewer claims stuck in substantiation and lower support call volume.
Shop
+20%benefit utilization
10M+users
Supplemental benefits like OTC medication and healthy food funds only help if people can use them, and many members couldn't. Most are older, many have low tech literacy, and they face digital and physical accessibility barriers. When benefits go unused, contract renewals are at risk and client funding can shrink.
Building on UX research, I led discovery and the build of a new online shop integrated with Amazon and Walmart fulfillment. The experience is designed around the member's balance: they always see how much they have left to spend, benefits apply automatically as items go into the cart, and a secondary payment covers anything over the balance.
- Benefit utilization up 20% with the new shop.
- Now the go-forward shopping experience for 10M+ users, reached through a structured migration.
- Amazon and Walmart fulfillment for eligible OTC and healthy food products.
Wanderwise
45.5%lower AI cost
4AI agents
36/36evals passed
4 guardrail layers
4 guardrail layers
Try Wanderwise ↗
Live flight search · sample hotels and activities · nothing is booked
AI trip planners fall apart on the details: budgets that don't add up, hotels that don't exist, double-booked days. I designed and built Wanderwise to draw a clear line: four narrowly scoped agents handle judgment, like interpreting requests and writing the itinerary, while tested code owns budget math, feasibility, inventory, and every change to the trip. I took it from brief to live app in a week.
The key product call was deciding what the model should never do. Every flight, hotel, and activity an agent suggests is checked against real search results before it's saved, so it can't invent inventory. When one-shot itinerary generation couldn't handle "swap the hotel" or "make it cheaper," I redesigned it into a step-by-step flow where each piece is proposed and confirmed on its own.
- Cut AI cost 45.5% with prompt caching, and measured no significant increase in response time.
- 36 evals run against the real AI and database, not mocks, including attacks like invented flight IDs and a planted prompt injection. They caught over-budget trips slipping through, which I fixed with an explicit override step.
- 4 guardrail layers: input and scope checks, validation of every AI output, business rules like budget and dates, and workflow permissions.
- Integrated live flight search through a third-party API (SerpAPI), alongside Next.js, Supabase with pgvector, Claude, and Voyage embeddings, with separate dashboards for AI cost and product metrics.
Off the clock
Personal
Tennis
Obsessed with watching, playing, and talking about tennis.
Cooking
From learning my grandmother's recipes to making multicultural fusion food, I wind down in the kitchen.
Mocha
Spending time with my dog, Mocha, is always a great way to stay present.
Get in touch
Let's talk
Open to new opportunities and interesting conversations. Reach out anytime.