Rohan Ramanath

Machine Learning | Product | Co-Founder, Hyperplane

[David Velez @velez_david]

Just released our “Data Nubank” report. The numbers on Brazil:

→ 115M customers — 60%+ of the population, the largest privately-owned financial institution in the country
→ 30% of Brazil uses us as their Primary Bank Account — the only fintech in the world to become #1 in a country
→ In 100% of Brazil’s 5,571 municipalities, including the 49% with no physical branches
→ 31.5M people entered the financial system for the first time via Nubank
→ R$134.7B saved for customers in fees
→ ~6M businesses, the largest CNPJ base in Brazil
13 years in, and it’s still Day 1.
https://t.co/HW0ZtJ45aD

[Eun Ro @Eunjung_Ro]

The AI transition is coming - here's how I'm responding.

With warnings of mass job displacement due to AI, I've been deeply concerned about the future workforce our children’s generation will enter. As this transition looms, we must devote more attention, experimentation, and commitment to supporting the human side of this equation.

That’s why we started @Go_JulyAI in late 2024: Reinventing economic opportunity for humans in the AI era. We are lucky to have begun our journey with trusted investors and advisors including @BasisSet, @Liquid2V, @SVAngel, @capital_olive, Night Capital, and Applied Intuition CEO @Qasar just to name a few.

I’ve shared more about how we will accomplish our mission at July AI here: https://t.co/ZTNTirgbOO. I would love to hear your thoughts.

[Jake @iamjakestream]

It is actually *easier* to work out every day than to work out 3x/week

[Rohan Ramanath @rohanramanath]

@Shotz018 @maiab 😂

[Kiran Bedi @thekiranbedi]

India makes it to finals. Have a look. #OlympicGames

[Alex Johnson @AlexH_Johnson]

7 months after raising a $6M seed round, AI startup Hyperplane was acquired by Nubank for an undisclosed sum.

Nubank is usually 12 steps ahead of everyone, so it's worth trying to understand exactly what they do and why it's valuable to Nubank.

Here's what you need to know 👇

- Hyperplane builds foundation models, similar to those built by OpenAI and Anthropic. The difference is that Hyperplane’s models are built on the much smaller first-party datasets of individual banks, rather than datasets scraped from the entire internet.

- The reason for this approach is that Hyperplane isn’t interested in generative AI for its content generation capabilities (which require massive multimodal datasets). It’s interested in generative AI for its ability to efficiently find predictive patterns within unstructured datasets.

- The genius of Hyperplane confining themselves to the unstructured datasets of individual banks is that it sidesteps almost all of the big regulatory and risk management concerns with generative AI in financial services. You’re only using your own data for training, so no need to defend to regulators why you’re using data from random subreddits to power your wealth advisor chatbot. And the outputs of Hyperplane’s foundation models are predictive insights and attributes, which can be used to build standard ML models and decisioning rulesets. This completely eliminates the risk of hallucination and makes model governance and explainability much easier.

To be honest, I’m a bit surprised that Hyperplane agreed to the acquisition. The terms of the deal were not disclosed, so it’s hard to judge from the outside.

What I can say is that every big bank in the world is going to want a Hyperplane at some point.

Nubank just got theirs and made it harder for their competitors to follow suit.

Smart.

[This Week in Fintech @twifintech]

Brazilian neobanking giant @nubank has acquired vertical AI for banking provider @hyperplaneai, which just raised its seed round 7 months ago.

(h/t @sytaylor for spotting this one)

https://t.co/W6zWRPjb1t

[TechCrunch @TechCrunch]

Nubank acquires AI-for-banks startup Hyperplane https://t.co/tzKki0koSd

[Andrew Escobar @andrewe]

Finally, something on the impact of AI + banking that doesn’t sound handwavy.

[Alex Johnson @AlexH_Johnson]

This was *exactly* the goal of my conversation with Rohan.

What are LLMs uniquely good at? What are their weaknesses? And where are the highest value areas inside of banks that they can be deployed in a compliant, low-risk way?

This is the podcast you’re looking for.

[Rohan Ramanath @rohanramanath]

I had an an engaging conversation with my friend @AlexH_Johnson on @FintechTakes. We walk through large language models (LLMs), constraints in personalization and predictive applications, and AI's ability to reshape banking.

https://t.co/xWqHdlq8Nq

[Rohan Ramanath @rohanramanath]

@pitdesi My co-founder @felipebpm is part of the most recent batch of Thiel Fellows!

Super proud of him 😍

[Rohan Ramanath @rohanramanath]

@dchaplot @GuillaumeLample congratulations folks!

[Jason Yeh @jayyeh]

1/ I hosted a live workshop last Thursday for 60 founders in Brazil

It was fantastic...and just a hint at what i'll be doing more of all around the world 🌎!

[Rohan Ramanath @rohanramanath]

@_arohan_ We recently published LAWN, which stacks onto most base optimizers. The LAWN variant of Adam improves generalization performance, even at extremely large batch sizes!

Link to paper - https://t.co/tLUhLpDHFt

[Rohan Ramanath @rohanramanath]

we'd love to hear your thoughts on our work
@aaron_defazio @MaxMa1987

[Aman Gupta @aman2304]

Introducing LAWN, a new family of optimizers for deep learning. LAWN can be added to most base optimizers like Adam and SGD to improve generalization performance, even at extremely large batch sizes!

Link to paper - https://t.co/yREFAlPkyQ

[Aman Gupta @aman2304]

Introducing LAWN, a new family of optimizers for deep learning. LAWN can be added to most base optimizers like Adam and SGD to improve generalization performance, even at extremely large batch sizes!

Link to paper - https://t.co/yREFAlPkyQ

[Barack Obama @BarackObama]

Four more years.

[Harsha Bhogle @bhogleharsha]

such supreme skill, such supreme athletes. but eventually the human body has to give.