Applied AI / Data Science

Scope

The mission for the role:

Popsa's flagship is Memories: we take a user's entire photo library and turn it into albums worth printing. Almost all of that is powered by machine learning we build ourselves: computer vision models that run on device across millions of phones, curation pipelines that work out what an album is, and LLM features that name and caption it.

We're hiring an experienced individual to own whole tracks of that work. The role title is flexible, this opportunity is about what you bring, your skills and talent, and passion for shaping an opportunity to make it your own. You'll take problems that arrive as a sentence ("our explicit-content filter isn't good enough", "we want pet memories") and turn them into shipped models: scoping the problem, defining what good looks like, training and evaluating the model, getting it onto device or into our FastAPI services, and keeping product and leadership up to date while you do it.

We're office-first, in our Soho office a minimum of three days a week, because a lot of a small team's coordination happens by being in the room. We use AI coding tools heavily and expect you to, but we also expect you to be able to explain every design decision and line you ship. Work is tracked in Linear in tickets small enough to move weekly, with progress written up as you go. By six months we'd expect you to have taken at least one project from an undefined brief all the way to production this way, having scoped it and defined its evaluation yourself.

What you'll work on

• Train, distil and quantise the computer vision models that run on device inside the app: content safety, aesthetics, object and scene classification, and new capabilities such as pet detection

• Build the curation pipelines behind Memories: detecting events, people and themes across a whole photo library and generating albums at scale

• Take large training runs from experiment to production on AWS: SageMaker training jobs on hundreds of millions of images, evaluation pipelines, experiment tracking, and the serving infrastructure behind our FastAPI services

• Integrate LLMs and vision-language models into customer-facing features such as title suggestions and captions and, longer term, on-device LLM inference

• Work directly with our iOS and Android engineers to get models running fast and small on real devices

About Popsa

Right now is one of the most exciting moments to join Popsa. Deloitte named us one of the UK’s fastest-growing technology companies, and the Financial Times recognised Popsa among the Top 5 fastest-growing software companies in Europe.

We’re backed by world-class investors, and the world is noticing. Our iOS and Android apps are available in 12 languages, trusted by more than 10 million people who have created over 5 million photobooks shipped to 50+ countries. With more than 350,000 five-star reviews, we’re one of the highest-rated consumer apps anywhere.

But the real story isn’t the numbers and it’s the memories behind them.

We’re living in a time where we capture more photos than ever before. Billions of moments sitting unseen in camera rolls and clouds, mixed with screenshots, receipts, or where we parked the car. We are great at documenting life… but not so great at preserving the meaning in it.

Popsa is changing that.

Since launching in 2016, we’ve built an award-winning platform that removes every barrier to transforming your favourite memories into something beautifully tangible. No design skills. No time-sink. Just the joy of holding real stories in your hands. A celebration of the people and moments that make life worth living.

✨ This is only the beginning.

👉 About us

👉 Read more about our journey so far

👉 Take a look inside our Soho HQ

What we’re building next

Today, we’re best known for photobooks, but our ambition goes far beyond printing. Popsa is creating a new generation of personalised experiences that combine advanced AI, thoughtful design and behavioural science.

Our goal? To help people emotionally process, reflect and connect with the meaningful events and relationships in their lives. To ensure memories don’t fade, but flourish.

From smarter storytelling tools to products that support wellbeing and strengthen bonds across the world, we’re reimagining how humanity remembers.

👉 Explore our vision for the future of Popsa

Private by design. We’re deliberate about what is processed on-device, what is uploaded, and how customer content is protected, so people can trust us with their most personal memories. 

Key responsibilities

• Own projects end to end: take an undefined brief, break it into pieces that can each ship inside a week or two, sequence them, and drive them through training, evaluation, deployment and monitoring without being handed a plan

• Define what good looks like before you build: the metric, the evaluation set and the false-positive bar the business will accept, agreed with stakeholders up front

• Understand before you build: ask the questions and read the existing code first, so that constraints surface on day one rather than in review

• Design, train and deploy deep learning models that reach customers every day, on device and on server

• Keep your work visible: short written updates, tickets that move, and interesting intermediate results shared as you go, so that product, engineering and leadership always know where a project stands

• Present results to technical and non-technical audiences and land on a recommendation, not a list of options

• Improve our infrastructure for large-scale training, evaluation and experiment tracking

• Work closely with product, mobile engineering and backend to ship features, and be the person the rest of the team learns from on training, evaluation and deployment

• Stay current on machine learning research and tooling and bring what's useful back to the team

Skills & experience

• Hands-on experience training deep learning models on large image datasets, including transfer learning and fine-tuning, and ideally distillation and quantisation for on-device deployment

• Track record of owning machine learning projects end to end in production, from ambiguous brief to deployed model, and of scoping and sequencing that work yourself

• Rigorous approach to evaluation: you can talk concretely about metrics you chose, evaluation sets you built, and thresholds you set, and why

• Hands-on experience running training and inference on AWS (SageMaker, S3, ECS or similar)

• Excellent Python, and the ability to read, explain and extend an existing codebase before changing it

• Docker and CI/CD as part of your normal workflow; familiarity with infrastructure as code (we use Terraform)

• Clear written communication: short design notes, headline-first updates, and PR descriptions a busy reader can act on

• Strong grounding in statistics and machine learning fundamentals

Any of the below would be ideal, but none are essential:

• Shipping models on device and working with mobile engineers on latency and bundle size

• Integrating LLMs or vision-language models into product features

• Large-scale training pipelines: distributed training, GPU scaling, high-throughput data loading

Personal attributes

A self starter, who is motivated to innovate with a growth mind set.