Tell Us a Bit About Your Role
I'm a Senior Data Scientist at Cleo. I manage a team of data scientists and analysts.
My main job is working on the machine learning models that drive many of Cleo's features. We’ve got a range of models, from NLP models for our chat experience, to models that understand user finances. We also have lending and credit risk models that are core to our salary advance offering.
How Did You First Get Into Data Science?
I studied mathematics at uni, and started my career in traditional finance on a trading floor, initially as a quant then later transitioning to trading.
After that, I took an extended sabbatical–meant for traveling–but ended up spending most of my time shuttling between two of my favorite places: Tokyo and San Francisco. It was at this time that I got into machine learning through personal projects and a couple of short courses. I was hooked.
Recognizing how exciting the growing field of machine learning was, I decided to go back to school and get a Masters in machine learning and data science.
After this, I wanted to get into the startup world. I spent a year at a very early stage startup, which was a fun whirlwind, then a year later joined Cleo around the time of series B. It’s been a fun journey so far 🚀
What's Your Favorite Project You've Worked On?
Some of the most impactful work I led was during the early days of Cleo's Cash Advance product, on the classifier that underpins it.
It was clear this was going to become important for the company, and it needed a lot of love. Through some painstaking iteration and improvements, and collaborating with new teams that were springing up around it, we managed to get the lending decisioning into a much healthier shape. Now, it's a major part of Cleo's offerings, and while it has grown and evolved, at its heart it's still built on the foundations which we laid down early on, which I take pride in.
More recently, I’ve moved to work on the models that power our chat experience. We’re building an AI assistant that truly understands each user’s needs and preferences, powered by LLMs.
This requires us to build the next generation of conversational interfaces, moving away from ‘feature search’ type to being meaningfully conversational. We want to build products that users love to interact with, so we need to understand what a user actually wants and allow the assistant to recommend and work with other models to take actions to deliver it.
This means we deal with all the fun and games that come with fine-tuning LLMs for various tasks and using the latest APIs in production. We’re dealing with the problems of the moment, for example:
- How do you work with these models to preserve Cleo’s unique tone of voice?
- How do you evaluate outputs meaningfully at scale both outright and in a comparative way?
- How do you combine the fluent generative output of a LLM with supporting models to figure out what a user wants to do and to give them ways to achieve it in the app?
- Which additional tools do we need to build so we can leverage the power of LLMs efficiently and effectively?
These are all fun and challenging problems. We’re moving fast and learning about the best way to solve them, for real, in a live app with millions of users, so it’s important we deliver a meaningful experience.
What Do You Like Most about Working in the Data Science Team at Cleo?
Cleo has a straightforward, transparent and low-politics culture. It's a place where proactive people can thrive. Here, you can quickly move from idea to production. Once we align on something, we get that MVP (minimum viable product) out as fast as possible.
We’re pragmatic in terms of organizing ourselves and our models. For our important business critical models, we will spend more time on versioning, AB testing and calibrating. Whereas for a fun new feature where we don’t mind too much about a false positive, we’re happy to roll out a quick locally-trained model and see what it can do in the real world.
We’re still on an exciting journey; I joined when there were just 4 data scientists and we did all the deployment work and the product analytics too in our ‘spare time’. Model release time could be tense…
Since then, we’ve grown and built out teams. We now have multiple product analysts, analytics engineers and data engineers, so we can go into much more depth in each of these areas. We’re still changing, and the Cleo you might join today will again be shaped by you and what you bring to the company, and be different in another year or two.
Where Can We Find You When You're Not at Work?
Is this a trick question? I have 2 toddler aged kids...
I suppose I might be listening to jazz, in the gym, also trying to read a Japanese book on my phone. Basically doing all my hobbies at once in my brief time off from them 😅
Any Advice for Potential Applicants?
We value self-starters who can create useful products in a world of imperfect data.
We also care a lot about impact. Even if you are deep in the modelling, it’s important to know why you are doing what you are doing and what it will enable, as this helps you prioritize and pivot when necessary.
We value product thinking in all roles, so an understanding of the kind of overall metrics the business would be interested in and how the machine learning models underpin that would also be useful to mention in an interview.