Ai Priority Focus started in 2019 when our founder, Callum Mercer, left a data-science role at a large UK bank. He had spent four years building models that looked impressive in Jupyter notebooks but never reached production. The gap between proof-of-concept and deployed system frustrated him enough to start a consultancy focused entirely on shipping.
The first project was a demand-forecasting tool for a regional food distributor. It ran on a single virtual machine, retrained itself weekly from a cron job, and saved the client about £6,000 a month in wasted stock. Nothing glamorous, but it worked. That attitude stuck.
By 2021 the team had grown to five. We moved into an office at United Kingdom, England, Sauer-under-Schmidt, BA00 0BX, 5 Ronald Close, close enough to the M4 corridor that we can visit London clients in under two hours. Today we are nine people: six engineers, one designer, one project manager and Callum.
We have no plans to become a 200-person agency. Staying small means every project gets senior attention. Callum still reviews pull requests. Our project manager, Rina, talks to every client at least once a week. That is the arrangement we like, and it seems to work for the companies that hire us.
These are not motivational posters on the wall. They are the rules we use when making trade-offs on a project.
A model in a notebook is a research exercise. We measure success by whether the system is running in production and producing value. Every project plan includes a deployment date, and we hold ourselves to it.
We quote a fixed price after the data audit. If the scope changes, we renegotiate openly. You will never get an invoice larger than the one you approved.
We write documentation aimed at the people who will maintain the system after we leave. If your analysts cannot explain what the model does, we have failed.
Accuracy, latency, drift, cost per prediction. We instrument every deployment so you can see exactly how the system performs over time, not just on launch day.
We are a compact group. Everyone listed here works on client projects directly; we have no layer of account managers between you and the engineers.
Founder and lead engineer
Former data scientist at Barclays. Callum designs system architecture, reviews all model code, and handles the initial scoping call for new clients. He has a particular interest in time-series forecasting and anomaly detection.
Project manager
Rina ran software delivery teams at Capgemini for six years before joining us. She manages timelines, client communication and the review-queue process for document intelligence projects. Clients speak to her weekly, sometimes daily during deployment sprints.
Machine learning engineer
Tom writes most of our NLP pipelines. He built the multilingual sentiment model for Kindwell Cosmetics and the retrieval-augmented chatbot for Greenfield Broadband. Before joining, he completed a PhD in computational linguistics at Edinburgh.
The rest of the team includes three backend engineers (Python, Go), a front-end developer who builds the dashboards our clients use daily, and a part-time DevOps contractor who manages our Kubernetes clusters. We are currently hiring a second ML engineer; if you are interested, email [email protected] with your CV and a link to something you have built.
AI moves quickly. A technique that was novel in January can be commodity tooling by June. We stay current through a few deliberate habits.
Every Friday afternoon is reserved for internal research. Engineers pick a paper, a new library or a dataset and spend three hours experimenting. Findings go into a shared wiki. About one in five experiments has led to a technique we later used on a client project.
We attend PyCon UK and the London Machine Learning Meetup regularly. Callum spoke at MLOps Community London in March 2024 about drift detection strategies for tabular data. The talk covered the monitoring framework we use internally, which is now open-sourced on our GitHub.
Certifications matter less to us than shipped projects, but for transparency: four team members hold AWS Machine Learning Specialty certification, and two have completed the Google Professional Machine Learning Engineer exam. We run most workloads on AWS, though we have deployed on Azure and GCP when a client's existing infrastructure required it.
If you want to see whether we are the right fit, the fastest route is a 30-minute discovery call. Ring us on +44 7389 684113 or fill in the form on our home page.