We help add high quality AI to your startup software quicker

What we specialize in

Integrate AI into your product

You have an existing product and are looking for opportunities to improve it using the latest AI advancements, whether it’s by making it faster for your users to achieve their goals or by scaling existing expensive processes. We can help you decide which problems to focus on first, define the user experience and the success criteria, as well as choose key technologies. The answer might not necessarily be an LLM - in many cases more lightweight models end up being a great fit and we’ll help you explore them.

Evals during development and in production

It simple to slap a call to ChatGPT into your product and call it done. But how do you evaluate the quality of the response and how it’s performing over time? Every time your developers update a prompt you risk breaking long-tail use cases. Third party models are continually changing too, so even if you don’t update anything, what you get back from them might - causing previously good result to become worse.

Whilst it’s relatively straightforward to check that inputs you and your team can think of are working as expected, how does that compare to what real people are using in your system.

We can work with you to define evals and processes for continuously monitoring the quality of the responses you get back.

Level up your AI: use your data

How do you stop being a ChatGPT wrapper and differentiate your offering? You need to leverage your data, which means you need to have it process it so it can be used for RAG / embeddings, fine tuning models, or building models from scratch.

We can assist with building the pipelines and workflows to make your data work for you to enhance your AI systems.

Compliance

People are increasingly concerned with the implications of AI and the data used to train and query it. We can help you identify the general and specific risks you are taking on by introducing AI into your products.

How we work with you

Our typical engagements look like this. We find this structured approach maximizes the success of AI projects by clearly defining the goals and risks and communicating them.

1. Discovery

First we'll meet with you to discuss what problems you want to solve and what the context is - who is your team, what is your product and data architecture, what data do you have access to.

We do this to make sure that at the end of the project you achieve your goals.

2. Planning

From this discussion, we'll create a design and project plan which will describe what the solution will look like and the milestones for delivery.

We will consider your team's structure, working style, and processes so we can be most effective, and we will design the solution so it integrates into your existing tech stack.

3. Development

After this, we will work with you to deliver on the plan. We usually work following agile principles, delivering often and seeking feedback. But we will adjust to how your team works.

We will communicate with you however you prefer - be it standups, slack, email, pull requests, or JIRA tickets.

We do this so you know what the status of the project is and you can react to what we learn - to ensure that what we deliver is achieving the goal.

4. Integration

We'll make sure the deliverable works within the constraints of your system.

We do this to make sure that, for example, the speed is adequate when part of the whole system.

5. Handover

We'll document and walk you through how the system works, so your team know how it works and can support it.

Who we are

We have almost three decades of collective experience adding AI into software. We’ve worked on all aspects of AI systems; collecting, cleaning, transforming big data, performing statistical analysis to generate insights, training models from scratch, fine-tuning LLMs, integrating existing LLMs in products, monitoring and assessing models in production. We can help you understand your problems, design solutions for them, and plan how to incorporate those solutions into your teams and processes.