Rav
Co-founder & CEO
2x startup founder and former EMEA Lead at Meta, where his team helped roll out new products and scale engagement. Previously built the "Netflix for News", where he brokered deals with the world's leading media companies.
Catapult deeply understands you and the complete context of your work. It already knows what you're working on, the backstory behind it, and what needs to happen next. As you work, Catapult continually learns, remembers, and improves, taking more meaningful work off your plate.
We're still early, with so much to do. But Catapult already outperforms Claude Fable 5 and comparable OpenAI models on most knowledge work tasks when connected to Gmail, Drive, Notion, Granola, and Slack. And it does so at a fraction of the cost.
We're a deliberately small, obsessed team of repeat founders, researchers, and engineers from Meta, Amazon Research, G-Research, Oxford, Cambridge, and Imperial.
Many seemingly impossible problems still stand between us and transforming how the world works. If that excites you, join us.
See open rolesWhat you'll help us solve
Every day, we solve problems at the frontier of AI, many of which look impossible. Here are just three examples:
How do we build an integrated understanding of all a user's data to create the most powerful context engine imaginable?
We don't use MCPs to pull in user information at inference time. We continually ingest data, much like Google does with the internet. We use sophisticated multimodal embeddings, alongside a host of other algorithms and tools, to search and cross-reference across sources. We build memory systems, learn abstract patterns, and construct high-level theories of the user and their data. And that's just the beginning. No one goes deeper than us, and it shows.
How do we build the most powerful personal coding agents?
A user's data is complex. It spans sophisticated sources such as databases, spreadsheets, CRMs, and many others that cannot simply be "pulled into context." These sources need to be understood abstractly and in relation to everything else known about the user, so our system can intelligently and safely write and execute programs to read from them and take action. How do we build that bespoke understanding? And how do we use it to create the smartest coding agents? It's an incredibly hard problem, and Anthropic and others have been struggling with it for good reason.
How do we manage long-running, valuable tasks that require enormous amounts of data, synthesis, and work: the long-context problem?
This famous open problem, with billions invested in solving it, comes down to building highly effective compaction systems. These systems mimic human memory, predicting what can be forgotten during a conversation and what will matter later. It's an enormously complex process in humans and remains unsolved across AI. Our end-to-end approach, particularly our context engine, opens up unique and promising ways to tackle it.
Curious about everything we've tried along the way that didn't work? Check this out.
Meet our team
Greg
Co-founder & CTO
Former Cambridge researcher, PhD student and CS teacher who taught mathematics, linguistics, statistics, theoretical CS, and AI. Worked on numerous topics at the cutting edge of mathematics and theoretical CS motivated by the P vs NP problem.
Mayeul
Founding Senior Engineer
Previously co-founded a deep tech startup where he built a quantum number generator. Helped build the computational teams at Imperial College London and UCL. Holds a PhD in particle physics.
Eric
Founding AI Research Engineer
Oxford and Imperial educated researcher who continued on to Amazon Research and JP Morgan. A specialist in reinforcement learning, he built an internal coding assistant using very small models while at JP Morgan.
Pierre
Founding AI Research Engineer
Educated at ENSAE Paris and Paris Diderot University, he has worked across quantitative finance, payments, and applied AI in both New York and London. His background includes building trading signals and NLP systems at G-Research, as well as developing Checkout.com's first machine learning fraud detection platform.
You?
Apply to one of our open roles below.
Why you should not join Catapult
We care deeply about the culture we're building. A culture where exceptional people can do the best work of their lives, have fun doing it, and solve problems that genuinely matter. That requires making deliberate choices about how we work. And it means Catapult won't be for everyone.
- If solving something seemingly impossible scares you rather than excites you, we're the wrong place for you. At Catapult we frequently deal with problems that no one (not Google, Anthropic, OpenAI, or the research community) has a handle on, and tackle them head on with enthusiasm.
- This is not a nine-to-five job. We're obsessed with the problem we're solving and determined to build the best product in the world. Nobody tracks your hours, but everyone thinks deeply about the work and takes real ownership of the outcome.
- We work together in person from our office in London. We believe difficult, ambiguous problems are solved faster when you can talk directly, use a whiteboard and keep moving. We are flexible for genuine personal needs, but we are not a remote or hybrid company.
- We have no room for ego. We care about finding the right answer, not defending our own ideas or proving who is smartest. We make mistakes, challenge each other directly and change direction quickly when the evidence says we should.
If you're more excited after reading the above, we want to hear from you.
Open roles
- Founding Engineer Engineering · London Apply
- Founding AI Research Engineer Research · London Apply
- Founding Operator/Chief of Staff Operations · London/San Francisco Apply
- Open application Don't see the right role? Shoot your shot. Apply
- Refer a candidate We'll give you $10,000 if we hire them. Refer
FAQs
How will you beat Anthropic and OpenAI?
Claude and OpenAI are building increasingly intelligent general models. We are solving a different problem: giving AI a deep, persistent understanding of your work.
Their connectors and MCPs retrieve a few emails, messages or documents after you ask a question. Catapult already understands how those conversations, people, projects and decisions connect, and keeps that understanding current as your work changes.
Models will keep improving and we can use whichever one is best. Our advantage compounds separately: a living model of you, your company and how work actually gets done. Better intelligence helps everyone. Better context makes Catapult better for our users.
What makes Catapult technically difficult?
Real work is messy. A single decision might begin in a meeting, continue in Slack, appear in an email and eventually change a document or customer plan.
Catapult must understand those connections across millions of pieces of information, keep that understanding current, respect permissions and reliably select the right context for each task. We are also solving evaluation, safety, document understanding and agent execution across wildly different customer environments.
What is your interview process?
Our typical process is:
- A founder interview focused on your background, ambition and how you think.
- A follow-up conversation about problem-solving and judgment.
- A deeper session where you present something you know exceptionally well and work through a real Catapult problem on a whiteboard.
- A paid work trial in our office.
- References & offer extended.
We do not rely on generic coding tests. We care more about how you reason, communicate and work on a difficult problem with the team.
Do you sponsor visas?
Yes. We sponsor visas for exceptional people and already have the infrastructure to do it.
We care about finding the best person, not where they were born or which passport they hold. If there is a strong mutual fit, we will work with you on the appropriate UK visa route and cover the company side of the process.
What compensation and benefits do you offer?
We offer competitive salaries and meaningful stock options because we want everyone at Catapult to share in the upside they help create.
Our UK benefits include:
- Bupa private healthcare
- A wellbeing allowance
- A tax-efficient pension scheme
- 34 days of paid holiday, including public holidays
- A workplace nursery benefit
- All the equipment you need to do your best work
What should I expect in my first few months?
You'll contribute from day one and take ownership of meaningful problems within your first month, turning them into something users rely on. Expect direct support, fast feedback and broad responsibility, not layers of management or a narrow box.