Data Scientist (Exploration Team)

About CloudWalk:

CloudWalk:

• CloudWalk is an AI first company building its own technology to bring justice to the broken payment system in Brazil. We are building what one would call a “self-driving bank”.
• Some people say a company is a unicorn when it reaches a valuation of over 1 billion dollars. We are one of those companies. But we are also a much cooler type of unicorn: a freaking amazing beast that is incredibly rare, cool as hell and damn hard to catch.

The R&D team:

• We love data.
• We love living in a time when there’s access to an immense corpus of shared knowledge and incredible statistical and computational tools to extract meaningful information from the data.
• We love thinking in high dimensions.
• We like to explore before we exploit.
• We sprinkle sci-fi references in everything we do.

What the job entails:

• You will be part of an exploration team inside the AI department. We take a step back from the day-to-day urgencies and pursue ambitious projects with high impact potential.
• You will join our efforts to discover and refine transformer-based neural network architectures, data pipelines, pre-training objectives, and fine-tuning strategies that will power the next generation of machine-learning models for finance.
• You’ll have access to tons of data, but you’ll also be cursed with tons of noise. Everyday events generate a torrent of perfectly normal behavior. To find truly valuable insights, you will need to follow clues, spot patterns, ask sharp questions, wrestle with uncertainty, uncover the story hidden in the numbers, and turn raw data into knowledge. You’ll need to be a detective.
• You will stay on the cutting edge, diving into both seminal papers and the latest conference breakthroughs, distilling fresh academic insights into inspiration for practical experiments for our models.
• You will have the chance to experiment. Do you think that the hot new model from a newly published paper can be applied to some of our data? Let's try it! Do you think the atmospheric pressure, the migration pattern of birds or the number of capybaras living on the banks of the Pinheiros river are good predictors of credit card fraud or creditworthiness? Let's investigate the data! Do you think there's a different perspective to look at a certain problem that could solve it better or complement our current approach? Let's put that to the test!
• You will leverage ample compute resources to run ambitious training cycles, scale up ideas quickly, and iterate based on what you learned.
• You might have taken a bunch of courses and learned all about NNs, CNNs, RNNs, SVMs, CARTs, RFs, LDA, QDA, XGB, BERT, SD, GPT, and all that nice stuff. Those are all techniques to learn a mapping from X to Y. Here, you will also have to think long and hard about what X you'll be mapping to what Y.
• If you never allow yourself to fail, you will never allow yourself to take risks. The next big breakthrough will not come from someone who's playing it safe. Here, you'll be expected to experiment and you'll be allowed to fail, as long as you learn something. Over 99% of all species that ever existed on Earth are now extinct. But the remaining 1% would not have succeeded without them.
• Not going to lie: sometimes you will have to do things you might not particularly enjoy. There will be times when we'll be at war, when we'll be together in the trenches, fighting the enemy, shoulder to shoulder, doing whatever needs to be done.
• As a member of a fully remote and distributed team, you are expected to complete tasks autonomously, being highly collaborative and self-driven. We expect you to have the curiosity of a child with the responsibility of a grown-up.

Requirements:

• Deep ML intuition. You think naturally in embeddings, matrices, tensors, projections, gradients, and loss landscapes—and you can explain those ideas without a whiteboard meltdown.
• End-to-end modeling skill. From data wrangling to metric-driven deployment, you spot promising ML opportunities and turn them into working systems.
• Fluency in Python (and friends). PyTorch, TensorFlow, NumPy, pandas, SQL—you’re productive across the modern ML stack and pick up new tools quickly.
• Data-driven detective work. You’re comfortable sifting through terabytes of noisy data to surface the patterns that matter.
• Research mindset. Reading arXiv before breakfast, reproducing baselines, and adapting fresh papers to real problems feels like fun, not homework.
• Parallel-experiment discipline. You’re organized enough to run and track multiple training jobs at once without losing your mind (or your metrics).
• Clear communicator. You debate model choices and share experimental results—in English—across time zones. Portuguese is not required.

Recruiting process outline:

• Our selection method is simple but hard. If you pass, you are definitely smart.
• 1) Online technical assessment
• 2) Technical interview
• 3) Cultural interviews
• If you are not willing to do an online quiz, do not apply.

Originally posted on Himalayas