I work in the Founder’s Office at Mihuru in Mumbai. Before that, I worked on Google’s Pixel pricing through Smollan and on a hotel’s costs in Dharamshala.
- 5%One of three vendors quoted without GST. Every comparison against it was off.
- 1 of 12Zepto Mumbai slots in stock for a protein brand’s new launch. Blinkit had 8 of 8. Found unasked.
- 32People in 7 countries used the pricing pipeline I pitched and built
The Divine Hima · Dharamshala · 2026
Can he find the number that is wrong?
My family’s nine-room hotel and its restaurant in Dharamshala. I have run its numbers since June 2026. No analytics team, no data warehouse.
- 5%
One of three vendors printed prices without GST. The other two included it. Every comparison against that vendor was off by 5%.
- ₹139 to ₹113
Rebuilt the cost of all 31 dishes from 17 vendor bills and 106 line items. The average cost per dish moved from ₹139 estimated to ₹113 actual. Modelled contribution up 28%.
- 73 to 31
Cut the delivery menu from 73 items to 31 and repriced every survivor to a cost-derived target. Then checked every published price live against the plan. Exact match. The 31-item menu is live at these prices.
- 426 to 633
Organic search sessions a month, July against the 30 days to 4 October, leaving out one guide that spiked on a cricket match. The growth came from the travel guides I published. Brand traffic to the homepage fell over the same months.
- 21 to 62
Clicks through to the booking engine from organic search. Most of the jump came the week I found the Book Now button hidden on phones and swapped the room pages’ enquiry form for a button.
- ₹0
Direct bookings from all of it so far. That is the open problem.
Google via Smollan · Aug 2024 to Mar 2026
Can he build the fix?
20 months at Smollan, a retail consultancy, supporting Google’s APAC and EMEA retail team.
- 2 to 4 hours
Every week the analysis was assembled by hand: pull, clean, format.
- 1,000+ rows
I built a Python pipeline that pulls 1,000+ rows of device and pricing data, adds AI-generated callouts and updates the dashboards on its own. It replaced the manual work.
- 32 people, 7 countries
The dashboards were used by 32 people in 7 countries, including Mountain View, in the monthly Pixel pricing calls.
Also: quarterly intelligence decks from three-plus years of panel data across 33 APAC and EMEA markets.
Decisions
Does he have judgement?
Seven times I went against the obvious move. Once I was wrong.
The Divine Hima
Chose Cold Coffee over Veg Manchurian as the free add-on
Manchurian is a wok dish that competes for kitchen minutes with dine-in orders. Cold coffee is made in seconds. Kitchen time, not rupees, was the scarce resource.
The Divine Hima
Declined a ₹90,000 photo shoot and built an image pipeline
A ₹2,000 phone shoot and an image pipeline did the job.
The Divine Hima
Refused the redirect the hosting dashboard recommended
The site already forced HTTPS. On that stack the new rule would have looped forever and taken the admin panel down with the site.
LoadBuddy
Shipped LoadBuddy with one screen and refused the second feature
The problem is one forgotten decision a day. The best tools remove the thinking. They do not add to it.
SuperYou Shelf Read
Corrected my own SuperYou finding after a second data pull
Zepto hides some out-of-stock rows from search. A missing row is not a delisting.
Google via Smollan
Pitched the pricing pipeline before anyone asked for it
A working prototype gets a yes faster than a proposal, and I can improve it while it runs. Here it also answered the two questions I knew my manager would ask: is the data right, and does the output read the way Google’s decks read. She could check both on screen.
Google via Smollan
Built a pricing deck my manager liked and the client could not read
I rebuilt the aggregation around the price changes her team tracked. I had designed it for me. Now I find out how the reader already measures the thing before I design anything.
Edge vision test
Gated releases on the worst cell, not the average, and it blocked my own model
It scored 47% on one cell, the new part under lamp light. An average hides the one place a line breaks.
Built on my own time
Does he ship unasked?
SuperYou Shelf Read · Sep 2026
A shelf audit nobody asked for.
Four findings from two evenings of public data.

Edge vision test · Sep 2026
A factory camera that learns a new part in milliseconds.
A three day test on a phone, a steel tray and a cloud GPU.
Portfolio Monitor · demo mode
A monthly research note that writes and sends itself.

LoadBuddy
One screen. Green means push, orange means recover.
How I work
How does he work?
How I use AI. I use AI agents heavily to do the building. I do not use them to decide what is worth building. This site was built that way: I specified it, tested it and shipped it.
Writing
Essay
What 20 months of building pricing dashboards taught me: 7 lessons
LinkedIn article
The best investment research is the kind you never have to doLinkedIn article
India, Soft Power and the Evolving Battlefield of Diplomacy
Contact
Based in Mumbai.
Happy to talk about anything on this page.

