Ankisha

Experimenting, Building , Learning | Data Scientist at Angelone | IIT BHU

[Ankisha @ankishasharma22]

Created this app "Merge Magnet"
https://t.co/F64JPQxLye
Describe your skills or interests — the app surfaces open GitHub issues from curated repos (your favourite research labs) . Filter by org, bookmark what you'll ship next.

[Ankisha @ankishasharma22]

@Akshatinvest @Akshat_World @X please help this account

[Akshat Shrivastava @Akshatinvest]

I lost my account with over 440,000 followers. And, X hasn't been helpful so far.

My account @Akshat_World has been hacked.

My team has tried to coordinate with X support. But, it just keeps hitting a bot. We are clueless at this point how to approach this further.

It had taken me several years to build this account.
Not sure if I will ever get it back.

Just hoping that good folks would help me escalate this. @elonmusk @nikitabier

Thank you :)

[Fuli Luo @_LuoFuli]

Imagination is the ceiling of productivity in the new era.
Inspiring imagination is the core of management in the age of Claw.

[Ankisha @ankishasharma22]

@felixhhaas Ambassador

[Ankisha @ankishasharma22]

@_adishj MOSAIC

[Ankisha @ankishasharma22]

Just made a full #GaneshChaturthi awareness video… with ZERO camera & crew.
All AI + a little editing magic
Script → ChatGPT
video clips → @GeminiApp veo3
Voice & Music → @elevenlabs
Edit → iMovie

(the message is sustainable celebrations 🌱❤️)

p.s I am aware its not perfect but i still love it!

[Ankisha @ankishasharma22]

@himanshustwts @Singh1Aryan lol😂 & thanks for initiating sanity of this place🙏

[Ankisha @ankishasharma22]

Built something for AI folks!

I scraped job listings from YC startups to answer:
"What data skills are actually in demand right now?"
No more guessing — see the hottest tools & frameworks companies are hiring for today.

Used: @cursor_ai + @Supabase + @boltdotnew + @vercel + @OpenAI + Selenium (Firecrawl gave me pain 😅)

If you're a data pro (or aspiring one), this is for you
https://t.co/7ABfEh1iGb

#DataScience #ML #buildinpublic

[Ankisha @ankishasharma22]

@RichardSSutton Thank you for sharing sir!

[Michael Truell @mntruell]

Some of Cursor's early team

[Ankisha @ankishasharma22]

Was reading a doc listing the features of a robot — vision modules, audio capabilities, hardware specs
— a future where robots are productised like smartphones.

Imagine influencers reviewing robots:
“This one has LiDAR 6.0, stereo depth, runs on the new BioEdge chip, and speaks 14 languages.”

A whole marketplace for robots — domestic, industrial, personal wellness bots?

Curious if India could have a niche — maybe in material design, or building the world's best affordable vision stack.

Feels like the hardware + software convergence we saw with smartphones will repeat with humanoids.
Maybe I’m late to the party and this sounds obvious to many — but it just clicked for me today 😄
https://t.co/wdccgVT0l4

[Ankisha @ankishasharma22]

AB testing stories 3/10

A/B testing in the real world isn’t always clean.
We ran an experiment:
Does sending a weekend investment-awareness push notification boost session time?

Control = no notification
Treatment = users intended to receive notification via CleverTap.

We instrumented everything , including a key sanity check:
We found a red flag -----
In the treatment group, 30–40% never received the notification.

Why?
-> Throttling rules
-> User opt-outs
-> DND settings
-> Device uninstalls
-> CleverTap filters
That’s when it hit us:
We’re not measuring the impact of receiving a notification.
We’re measuring the impact of being assigned to one.
So we redefined:

Primary analysis = Intention-to-Treat (sent group vs control)
Secondary = Per-Protocol (only those who actually received vs matched controls)

Takeaway:
Treat ≠ Receive.
Good instrumentation + definition clarity = trustworthy results.
#ABTesting #Experimentation

[Ankisha @ankishasharma22]

AB testing stories 2/10

A few months back, we ran an experiment on ~8 million users. The goal? Measure the impact of communication across the board. We designed it well, analyzed the results, and moved on.

Fast forward: A new hypothesis came up —
"What if sending too many notifications to active users leads to fatigue and hurts experience?"
We started planning a fresh experiment: defined cohorts, treatment arms (1–2 vs 3–4 vs 4–5 notifications/day), set up metrics, MDE, the works.

Then came a lightbulb moment
"Wait, can we use the old experiment's data to validate the new hypothesis?"

Turns out, YES.
Even though we didn't have exact notification variations from the old test, we could slice the data cleverly to test a tangential hypothesis.
Result?
-> Early signals.
-> Strong guesswork.
->Saved weeks of setup, effort, and user fatigue.

Takeaway:
Good experiments are not one-time events.
If you design and log them smartly, they become data mines you can revisit to answer new questions without annoying new users.
Creative thinking turns past experiments into future superpowers.

#ABTesting #Experimentation

[Ankisha @ankishasharma22]

@saankhyaCFC8 True ! 🥲

[Ankisha @ankishasharma22]

AB testing stories 1/10
Everyone talks about the science of A/B testing.
But the real game? Operations at scale.
We ran an A/B experiment across millions of users:

Treatment: Send Push Notifications

Control: No Push Notifications

Simple, right? Nope.

Operational nightmare:
Push campaigns were managed by 9–10 different revenue teams.
We had to exclude the control users manually across all their workflows.
One miss? Leakage.

To monitor, some team members deliberately enrolled themselves into the control group — if they received any push, it meant leakage.
Day 1 & Day 2:
We caught leaks early.
Fixed broken pipelines.

Re-communicated exclusions to all campaign owners.

Lesson:
A/B experiments at scale are not just about statistical formulas, but also about process rigor, early detection, and alignment across teams.

Good experiments need good ops muscles too!

[Ankisha @ankishasharma22]

As we saw in the blume report, India 2 favors micropayments, willing to pay the same cost for same duration in smaller payments but unwilling to go for monthly subscriptions.

Maybe we just need to amplify the scale, on both time and money axis to observe the same pattern in India 1.

[Chirag Barjatya @chiragbarjatya]

When I got a dishwasher, I went to Reddit to read about it, and I was surprised to see that in the USA and the UK, people have been using this one of the best inventions of humankind for more than 30 years!

30 years. In India, it’s still not common to use the dishwasher.

- it cleans the best
- no holidays
- no fights of aaj bartan jyada hai
- it uses less water

Get a dishwasher please. Stop troubling members of your house to wash dishes.