I'm Sarthak, a product manager. I've built B2B platforms at Jio and Blinkit that were already live, so a bad release shows up in somebody's week. Now at IIM Bangalore.
Looking for a Summer 2027 internship in AI, commerce or platform products.
Three products that shipped and one that didn't. Open any of them for the full story.
Rebuilding the platform 10,000+ Blinkit sellers run their business on.
View projectHotel Wi-Fi an IT team can actually run.
View projectOne console for the five tools site managers were switching between.
View projectThe mentorship platform I co-founded, and what I learned when it stopped.
View projectI started in product at Jio, moved to Blinkit's SellerHub, and joined IIM Bangalore to get better at choosing what's worth building.
Coffee, music, cricket, and a Formula 1 team I've backed through good years and lean ones.
Where quick commerce seller tools go wrong on data, and where AI earns its place. September 2026.
Rebuilding the platform where Blinkit's sellers run their business, on three tracks at once.
The problems sat in different places. The flows were built for a desktop nobody was sitting at, everyone saw everything regardless of role, and the number a seller needed was always one export away.
We planned 24 features. After seller calls, session data and a shift in a dark store, the first release was six.
Every figure on this page is a movement against that track's pre-launch baseline. I can walk through the baseline and the window for any of them.
Whole platform, customer interactions and flows
I rebuilt the flows around what a seller is trying to finish and pushed for mobile first, since the work happens in a warehouse aisle. We shipped the mobile flows before the design system was done. Within a fortnight sellers showed us which flows they were abandoning, and those got rebuilt first.
Entry security, rebuilt as role-based access control
A warehouse operator and a finance lead were seeing the same screen. I treated that as a support problem first and a security one second, and scoped access to the job instead of the account. The narrower your role, the quieter the interface you get.
Every dot is a surface somebody can reach. Before, the list looked the same whoever logged in. Flip it and watch the map contract to the job.
Everyone saw everything, regardless of role.
| Role | Catalogue | Pricing | Inventory | Payouts | Reports | Users |
|---|---|---|---|---|---|---|
| Account owner | ||||||
| Ops manager | ||||||
| Finance | ||||||
| Warehouse | ||||||
| Support |
Illustrative roles. The production role map stays inside Blinkit.
Build the instrument, not the argument
The insights tab existed long before anyone opened it, because it sat outside the flow sellers were already in. I moved it inside that flow, and that's what got it used.
GRN checks meant sellers reading invoices line by line, with mismatches turning into payout disputes weeks later. I put OCR on the invoices so the check happens at receipt, which gave sellers back hundreds of manual hours a month.
Hotel Wi-Fi an IT team can actually run.
Hotel Wi-Fi fails quietly. A dead zone in the basement, a captive portal nobody can clear at check-in, and an IT team that hears about a dead access point from a guest rather than from a dashboard.
I started with the market, benchmarking 15+ competitors, and positioned JioWave as a premium offer at a price property owners could justify.
The build covered the whole loop: coverage planned off a heatmap, guest, staff and IoT traffic on separate VLANs, and a gateway that fails over to a second link before the front desk notices. Captive portal, policy, analytics and firmware updates all run from the cloud, so most fixes happen without sending an engineer to the site.
After the redesign of the operator flows, satisfaction went from 2.9 to 4.2 out of 5 and platform adoption grew 130%.
These are movements only, because each contract is a single named enterprise and an absolute figure would name the client.
An operations console for people who were switching between five tools.
A site manager answering one question was opening several tools. Is this property drawing too much power, which camera zone flagged an incident, which machine is about to fail, did that batch pass QA. Those questions belong together, so the redesign put them under one navigation and one visual language.
Four teams wanted their module on the landing screen, and each had a decent case. I spent most of a quarter working out what one operator needs when they open the console at nine in the morning. The landing screen kept three things: live alerts, the site map and plant availability. Every team's detail moved behind its own tab, and I held that line through every review.
Numbers are relative movements only, because each contract is a single named enterprise.
A mentorship platform I co-founded and wound up after about a year.
Students needed a credible mentor and didn't have the network to find one. We built the place they could find one, it got early traction, and then it stopped.
We built on assumptions instead of talking to students, and the mentor revenue share was never going to pay for itself. The signal we missed longest mattered most. Students wanted placements and projects they could point at, and we'd built for guidance. By the time we had those conversations there was a volunteer team, a live product and no runway to change direction.
I run those conversations first now.
Here is how each seat happened, newest first.
Doing the MBA to move from shipping features to deciding which ones get built.
ISB waitlisted me after a good interview, and I knew why. I had scored 725 on the GMAT while rebuilding the Managed Wi-Fi network at Jio, but my products had been niche, sold to enterprise customers who buy very differently from the customers a consumer company builds for.
So I wrote CAT as well and scored 99.79 percentile. I sat the interviews while working at Blinkit, learning more every week until my last day, because I wanted to get the most out of that job. I chose IIMB for the strategy side of product, the part the job had not yet taught me.
SellerHub revamp, across three verticals.
After ISB I stopped applying to schools and started applying to roles that fit. E-commerce and the logistics behind it had always interested me, so I switched. Blinkit came through.
In my notice period at Jio I read up on the industry: which patterns worked, and the reasons they worked.
Managed Wi-Fi and the Unified IoT platform.
The market had just crashed after the 2022 boom, and off-campus PM roles were not coming. I had chased product for two years while my friends got placed, and I nearly gave up. I sat for college placements and got hired as an SDE at Jio, in the graduate engineer trainee pool.
When Jio's teams were being allocated, I reached out to every VP who was hiring and showed them how much I wanted product. They moved me into the management trainee pool, which was meant for post-MBAs. Everyone in it had years more experience than I did. I caught up in about six months, took on the most ambitious projects, and delivered.
Electronics and Communication, with an economics minor.
I was surrounded by people who were better at coding than me and visibly happier doing it. That pushed me toward the work that sits between the technology, the business and the person using the thing.
A lot of the education happened in the fest office. When the senior organisers left mid-planning on a 25,000-person fest, I took over communications and the decision process without a title, and we ran it without a safety incident.
Two internships did the rest, one research and one corporate. Neither came with a PM title, and both are the reason I went looking for one.
Four moments, and the habit each one left me with.
Two days before the Unified IoT launch, the core server went down. I took all external communication myself and gave the client a recovery plan I could actually keep. Inside the team I split the problem by module so people stopped reading the same logs.
Shipped Unified IoT on the launch date.
The senior organisers left a 25,000-person fest halfway through planning. Nobody handed me the job. I set up who talks to whom, how decisions get made and who signs off, and we ran the fest without a safety incident.
Ran a 25,000-person fest with no title.
On a research internship at IIM Bangalore, the analysis rested on the government's daily COVID figures. I cross-checked them against obituary records. The two didn't match, and the gap became a finding in its own right.
Found a 45% discrepancy in the dataset.
When VikalpWeb plateaued, a teammate suggested listing big-name mentors who hadn't agreed yet. I said no. We fixed mentor onboarding, put up real student testimonials and ran campus drives instead. Growth stayed slower, and everything on the platform stayed true.
No borrowed names.
Coffee, music, cricket, and a Formula 1 team I've backed through good years and lean ones.
I've followed Red Bull through the championship years and the ones where the car isn't there. What keeps me watching is the engineering, where a team finds time inside regulations every rival has read too.
Fruity and light-roasted, iced more often than not. On the counter: Hario Switch and V60, a Timemore C2S, a Sipologe scale and a gooseneck kettle.
My top 15. Tick the ones you've had.
0of 15 you've had
These fifteen are the ones worth going back for. I keep the list current.
I played guitar through the Trinity grades, and that's where the collecting started. The wall runs from indie tracks sitting at 15,000 plays to the Bollywood soundtracks everyone grew up on.
I captained the cricket side at Ice Breakers, IIMB's intra-college tournament. Badminton is the one I still play weekly. The rest is cooking, and photographing wherever I travel.
Where quick commerce seller tools go wrong on data, and where AI earns its place.
The question for a seller tool shouldn't be how good the dashboard is. It should be whether the right number shows up at the moment a decision is being made.
Most seller platforms treat reporting as a destination. There's a tab, it has charts, and the team measures success by how many sellers open it. At SellerHub the insights tab existed for a long time before anyone used it, and adding charts wasn't what changed that. Moving it into the flows sellers were already in did.
I think that generalises, and quick commerce makes it sharper. A seller on a dark-store network makes small decisions all day: restock this SKU, fix that price, chase a payout that came in short. Each one takes seconds and happens inside a task. A report that lives on its own tab asks the seller to stop the task, go looking, and come back with an answer. Most don't.
So the right number has to show up where the decision happens. Stock cover belongs on the restock screen, margin impact on the price-edit field, and the reason for a payout gap next to the payout.
This is where AI is useful to a marketplace seller, and it's narrower than the pitch decks suggest. Nobody needs a chatbot that restates their sales. They need the system to notice the three things that changed this week, say why in a sentence, and put that sentence where the fix happens. Invoice OCR is a small version of this. It reads a document nobody wanted to read and flags the mismatch before it becomes a dispute.
It also changes what a PM should measure. Tab views and report downloads reward the wrong design. Time from a problem appearing to a seller acting on it is harder to instrument, and it's the number I'd want on the wall.
If I were building seller tooling today, I'd start by listing the ten decisions sellers make most often and checking which of them already have the data they need sitting next to them. On most platforms the honest count is low.