Fuzzy Labs

What they do

Fuzzy Labs builds and productionises AI models for enterprise clients, using open-source MLOps tooling throughout. The work spans the full journey from model development to deployment and monitoring — the team describes itself as an extension of a client's in-house data science function rather than an outside vendor parachuted in.

The consultancy is led by two co-founders: Tom Stockton (CEO), whose background is in DevOps, and Matt Squire (CTO), who drives technical direction. The wider team of around 20 people is heavily engineering-led, with several staff holding PhDs from institutions including the University of Manchester and Durham University. Specialisms across the team cover recurrent neural networks, ensemble learning, uncertainty quantification, explainability, recommender systems, search and large language models.

Case work listed on the site includes building an AI assistant that halved case file preparation time, onboarding new polytunnels in days rather than weeks, and stress-testing AI security tools to uncover critical flaws. The public sector is a named vertical: Jamie Smith, a former South Yorkshire Police officer with 15 years in digital, data and technology, leads policing work. Bilal Khan, a University of Leeds Computer Science graduate, has focused on machine learning and LLM projects in the energy and public sectors.

Fuzzy Labs also publishes a practical guide called Cooking with MLOps — described as a cookbook of templates and patterns for building AI systems — alongside a blog covering topics from agentic AI to open-source tooling. Blog output in 2025 includes pieces on measuring agent effectiveness, the cost of agentic SREs, and running an internal AI hackathon.

The company is registered at Heron House, 1 Lincoln Square, Manchester, and operates out of the GM Digital Security Hub (DiSH).

What sets them apart

The core argument Fuzzy Labs makes is straightforward: open-source MLOps tools give clients ownership of their solution, avoid vendor lock-in, and are more adaptable as requirements change. They are explicit that proprietary platforms — of which there are many in this market — tend to be inflexible and expensive over time.

This is not just a sales position. The team contributes to the open-source community and publishes resources publicly. CCO Robbie Vigers brings commercial experience from Peak AI, a Manchester-based AI platform company, which gives the leadership team a clear view of what the proprietary side of the market looks like from the inside.

The team's academic depth is notable for a consultancy of this size. Multiple engineers hold PhDs in physics, astrophysics, machine learning and computational biology — disciplines that tend to produce people who are comfortable with uncertainty and rigorous about measurement. That background shows up in the blog's focus on agent reliability and effectiveness metrics, topics that most AI consultancies are still hand-waving about.

Responsible AI is a stated concern, particularly in the policing and public sector work, where Jamie Smith's background shapes how deployments are scoped and governed.

What they offer

• MLOps consultancy — embedded engineering support to help data science teams build, deploy and monitor machine learning models using open-source tooling.
• AI model development — end-to-end build of production ML systems, including LLM and agentic applications, for enterprise and public sector clients.
• Model monitoring solutions — implementation of monitoring infrastructure to track model performance in production.
• Cooking with MLOps — a published practical guide (described as a cookbook) covering repeatable patterns and templates for building AI systems.
• Open-source projects — publicly available tools and documentation contributed to the wider MLOps community.