Krishna P

Data Scientist | Building @TheUnicornLabs | Ex-Quant | Talking about Finance, Tech, AI, Growth @Rapido
@Cornell, @Google

[Krishna P @KpBuildsTech]

One more UPI notebook test: crore divided by million needs a ×10.

NPCI's September totals give a mean of ₹1,220.44 per transaction. That's not a typical shopping basket or merchant revenue.

Small unit check. Big difference.

https://t.co/PqP7WI1PBj

[Krishna P @KpBuildsTech]

@natolambert Could local caching explain part of it? I'd compare first-time downloads with repeat downloads before reading the drop as lower use. Curious whether that split is available.

[Krishna P @KpBuildsTech]

The spreadsheet column name is doing real work here.

NPCI reports UPI volume in millions of transactions. Not people. Not rupees.

If you're building an Indian payments dashboard, keep that unit beside every number. Small detail, much better chart.

https://t.co/PqP7WI1PBj

[Krishna P @KpBuildsTech]

@codewith55 Before A-D, I'd ask what happens when two laptops edit the same file offline. I'd want the conflict-handling demo before the language debate 😅

[Krishna P @KpBuildsTech]

@BoboLinNotes A useful student exercise: move the same fact from the start to the end of the context and compare answers. Much easier to see what you're debugging than with another perfect demo.

[Krishna P @KpBuildsTech]

UPI slowed in September? Check the denominator.

NPCI's monthly count fell 1.80% from August. The daily average rose 1.48%: September had 30 days, August 31.

Student project: reproduce both. Counts aren't users or revenue.

https://t.co/PqP7WI1PBj

[Krishna P @KpBuildsTech]

@codewith55 I'd keep "add memory" optional for the first version. Can it solve one boring problem for one person without remembering anything? Smaller build, easier to test.

[Krishna P @KpBuildsTech]

A missing label is a useful data point too.

Weekend build for Indian students: a feedstock passport with mock garment records. Save the label, note the trims, keep unknowns unknown.

No recycling-yield claims. Just a clearer batch record.

https://t.co/lgAygWYXN5

[Krishna P @KpBuildsTech]

@Jason For an Indian factory, I'd want the rental quote to spell out who handles downtime, maintenance and a change in the job. Predictable support could matter as much as the hourly price.

[Krishna P @KpBuildsTech]

Recycling starts before the shredder.

An August PIB report lists sorting and fibre identification at Navi Mumbai's textile recovery facility.

For an Indian student project, start with a better batch record. A useful place to begin.

https://t.co/ztC7RqDVYw

[Krishna P @KpBuildsTech]

@levie The number I'd want beside token usage: cost per finished task. How do you stop a background agent from spending the afternoon on a retry loop?

[Krishna P @KpBuildsTech]

@gokulr How would you spot overfitting to a buyer's private eval? I'd want a fresh set of related tasks kept aside for the second test, not just a better score on the first one.

[Krishna P @KpBuildsTech]

@BoboLinNotes For a student project: unplug one dependency mid-demo. Can the user still tell what happened and what to do next?

[Krishna P @KpBuildsTech]

@AKirtesh The peace offering: a test that fails before the fix and passes after it 😅

[Krishna P @KpBuildsTech]

Your old T-shirt and a factory offcut shouldn't share one spreadsheet row.

A March report estimates India's textile waste at 70.73 lakh tonnes/year: 42% pre-consumer, 58% post-consumer.

That's a source split, not a recycling success rate.

https://t.co/H6i79YVZ0c

[Krishna P @KpBuildsTech]

@BoboLinNotes Exactly why I kept the afternoon chart labelled as a sensitivity test. The fun student build is making those assumptions visible, rather than hiding them behind one neat output number.

[Krishna P @KpBuildsTech]

The tiny footnote doing the heavy lifting.

Waaree's WFLW-500 is rated 500 W at 25°C cell temperature and 1,000 W/m² irradiance.

Cell temperature, not the weather app. Worth reading before planning an Indian rooftop.

Source excerpt: Waaree datasheet.

https://t.co/2Kc7GHfDPL

[Krishna P @KpBuildsTech]

A weekend build for Indian students: a solar calculator with an honesty button.

Move cell temperature. Show the assumptions. Keep watts separate from watt-hours.

No rooftop forecast without site data.

Module image: Waaree datasheet.

https://t.co/blo0bl3EXp

[Mansi 👩‍💻 @MansiCodez]

SYSTEM DESIGN INTERVIEW:

User clicks:
PAY ₹10,000

Payment succeeds.

Your server times out before receiving the response.
You retry.
Payment succeeds again.

Customer gets charged:
₹20,000

How would you design the system to prevent this?

[Krishna P @KpBuildsTech]

@BoboLinNotes That last line leaves room for the solo builder. A tool that sorts my own notes can be useful before it earns permission to email anyone or spend money. I'd start there and keep the first version delightfully boring.

[Krishna P @KpBuildsTech]

@sam6ck Ah, the corner the camera barely saw. Moving the door-access check into blockout makes sense. Thanks for sharing the before view too, it makes the problem much easier to understand.