Practical AI Consultancy

Tired of the HYPE?

Get straightforward advice ABOUT:

Defining your AI Strategy

Developing your own models and IP

Designing Safe and Reliable applications

Deploying and serving your models

Defining Your AI Strategy

Are you wrestling with any of the following?

  • How should my business be using AI? What are competitors and customers doing?

  • Is AI a threat to my business? Does it create opportunities?

  • Should we develop our own AI or use existing products and services?

  • How can we use AI to add value to our data?

Our AI Strategy service answers these questions.

What will I get out of it?

  • A shared understanding and consensus among your team.

  • A compelling story about how your business will use AI to thrive and compete.

  • A realistic and concrete roadmap.

  • Clear guidance for initiatives affecting products, people, processes or data.

  • A detailed plan for the “first step”, enabling you to begin implementation immediately.

Developing your Own Models and IP

Have you decided to develop your own AI solutions? Perhaps you need to keep your data private, or you want to develop your own, differentiated capability?

The greatest challenge today is finding a team who have done this before – not easy when the technologies are both new and evolving fast. Our team has practical experience in:

  • Annotation and enrichment projects to develop datasets for large scale machine learning.

  • "Fine-tuning" Generative AI models to create unique capabilities.

  • Developing cutting edge algorithms for prediction, text processing and image analysis.

Designing Safe and Reliable Applications

Worried about the risks involved in deploying AI?

Here are some of the concerns we hear:

  • Could the model produce inappropriate or innacurate content?

  • Could the model exfiltrate sensitive or secret data?

  • How can I ensure consistent, high-quality outputs?

We design and test AI-powered applications to minimize these risks. For example, by:

  • Pre-processing and cleaning data used for in-context learning.

  • An iterative approach to improving prompt design and response-templating.

  • Using a range of tools to probe for undesired behaviours, including: hallucinations, jailbreaks, toxicity, factualness and data leakage.

  • Production Guardrails to scan and correct model responses, for example to mask Personal Identifying Information or remove toxic language.

Deploying and Serving your Models

Bringing a data science team’s work to life is hard. Many businesses struggle to embed and exploit machine learning models.

As models get larger and more powerful, the challenge is getting harder.

Let us help you with your AIOps and MLOps. Let your team concentrate on developing the models and your users focus on using them.

Practical and expert Help with AI and MACHINE LEARNING

In recent years the world of technology has been moving at a frenzied pace. Amidst such change, it’s hard to find people who really understand the emerging technologies or who have mastered new, specialised domains. In particular, in the field of AI, the art of the possible is being re-written every few months.

We are a team of highly experienced software, data and machine learning professionals who specialise in modern AI. We know what works, what is possible and what isn’t.

OUR PROMISE

Our mission is to help you thrive in an era of accelerating technological change. We promise to provide clear-headed advice and pragmatic technical support. And if we don’t know, we’ll say “we don’t know”.

Our Leadership Team

Hover for alternate images, created using Envvi/Inkpunk-Diffusion and lllyasviel/ControlNet-v1-1

Will Hardman

Data Science Director

Will Hardman is a former Chief Data Scientist and an expert in Natural Language Processing and Deep Learning.

He has been working with Generative AI since long before the current explosion of interest, and with machine learning more generally for nearly 20 years.

Will is responsible for helping Veratai’s clients drive adoption of modern AI and data science practises.

Matt Jones

CEO

Matt Jones was formerly the COO and then CEO of a software engineering consultancy. 

He specialises in the successful delivery of complex data and software engineering projects.

Matt is responsible for ensuring that Veratai’s clients receive high quality solutions, on budget and on time.

Charlie Anderson

CTO

Charlie Anderson has worked as a Lead Engineer, a Software Architect and CTO. 

Charlie designs the platforms and software that turn concepts into fully fledged solutions for Veratai’s clients.

Our Team’s Current and Past Work

  • Artist's impression of an AI Strategy conference

    AI Strategy

    Our client sought to understand how AI technologies could be harnessed to increase the adoption, integration and value delivered by their mapker-leading data product.

    We integrated findings from interviews (staff, customer and wider ecosystem stakeholders) with deep industry knowledge and AI-specific research to develop a set of structured roadmaps and benefits cases for the high impact themes we identified.

  • Revenue Management

    We worked with a global ticketing platform to develop a pricing, forecasting and reporting system to optimize revenues and yields.

    The result was a bespoke analytical capability to which the client has attributed a 3% revenue uplift annually.

  • Airport

    Predicting flight times

    We developed algorithms to enable airports in Europe and the USA to accurately predict the turn-around and departure times of aircraft.

    The algorithms are used to enable more accurate scheduling and resource allocation, improving the rates of on-time departures.

  • Medical Entity Linking

    We helped a multinational healthcare organisation execute a machine learning challenge to advance the state-of-the-art in medical Entity Linking.

    The competition attracted over 500 participants. Winning entries included the first use of Generative AI models in the field.

    The project created what is now the largest, publicly available dataset of clinical annotations.

  • Artist's impression of a matching algorithm at work

    Insurance Matching

    We helped a global re-insurer automate the matching of prospects to know corporate entities.

    Our fuzzy matching system was capable of efficient “any-to-any” matching over datasets in excess of 100,000 records each.