The Virtual Forge's 3 Pillars of AI

‍Artificial Intelligence is not in any way a new thing. While it has evolved and continues to do so, it has twin origins in the increasing automation and data enablement of many organisations, and the flights of imagination of some of our greatest artists and thinkers.

At The Virtual Forge, we have developed as an organisation, following a lineage just as technology has changed - from using Adobe Flash, classic ASP, and vanilla JavaScript through to iPhone development, streaming development, .NET core, ReactJS, LAMBDAs, NoSQL and many others. You might or might not know what any of these technologies are, but they illustrate the changes that have come along in the fifteen years that The Virtual Forge has been in business.

AI is a natural progression of this. When you strip away all the hyperbole and clickbait, it is a development of the skills and ideas that we as a group of enthusiastic technologists have honed, developed, and worked to put at the service of our clients.

Because of that, we are excited to blend our best-of-breed practices with the latest and newest opportunities that Artificial Intelligence can provide. We approach it as we have always done, with an imaginative and critical thought process. Both processes are always directed to answering the same question: What can AI do for our clients?

So, let’s break it down. The Virtual Forge’s 3 Pillars of AI seek to simplify and help answer exactly that question in a pragmatic and business-focused way.

PILLAR 1: AI Enablement and Integration  

In this pillar, we focus on understanding what your use cases are for Artificial Intelligence. This can take several formats - some clients come to us with a very precise use case such as ‘We want it to help with speeding up our invoicing process,’ and we can help you understand what is possible and what is not. We can guide you through what the costs are likely to be and help you understand where your data will reside, and how it will be processed.

In some cases, we are asked to simply help explore what is possible, or to look for places where  AI could be useful. Our experts will help look for pragmatic ‘quick wins’ and more strategic ‘big hitters’ to help you build a business case and explain likely outcomes. 

We also have many years of working with IT and compliance departments in various large multinational organisations (see governance below). Our team of experts can help you understand the sort of questions and concerns they will have as well as build presentations and datasets to enable them to feel comfortable with what you want to achieve. And, of course, ensure that all due diligence is done.

Beyond the discussions, we can help you build a proof of concept, something to ‘put meat on the bones’. There are many flavors of AI integration work, all of which we have experience in, and can quickly develop solutions that show what can be done. We can take a tedious process and automate that, we can help use long-hidden data to answer real business questions, we can enable an AI agent to complete a task and take some autonomous actions based on different forms of information such as creating a booking, evaluating a response, driving a recommendation,  doing some research, or any one of a hundred different micro uses.

Sometimes we do this by helping integrate an off-the-shelf agent, sometimes by feeding data to an existing model, sometimes by building a custom agent, or sometimes by retraining a whole model. It all depends on being able to do what suits your use case and budget, and we are proud that we have the depth of experience and enthusiasm for the subject matter to always be well informed to guide our clients in this ever-changing area, 

Whether we are building the proof of concept, or taking the next stage and building the whole solution, we are always transparent in what we are doing and how we are doing it. Our team of experts utilises architectures based on cloud partners like AWS and Azure and makes informed choices about the Large Language Models (LLM) in use. Think of them like the brains of the systems, with the systems we integrate with and the code we write like the arms and legs. Not all LLMs are created equal and our team of experts are consistently experimenting and exploring the new models created. 

• What were they trained on? 
• Is your data being passed on (anonymized or not)? 
• What is the latency in a request? 
• How reliable is the output?

Often with an LLM, you will need to build a set of checks and balances around the answers, especially if you have sensitive audiences, or you want to ensure a consistent output, such as ensuring the same JSON is output to drive another part of the process. We can help with all this and explain what we are doing, as we do it.

Our aim in this pillar is to make AI available to our clients in as large or small amounts as they want it, making sure we are always explicit about the business value and transparent about what underpins what we have delivered.

PILLAR 2: AI Data Preparation and Management 

In this pillar, we focus on helping you make sure that your data is in a state that can be usefully and accurately consumed by the Large Language Model. Once again, this is just a development of the work we at The Virtual Forge have been doing for many, many years. Helping our clients get their data ready for analytics and data lakes.

The old days of industrialized ETL processes for structured data are largely fading and the hunger for data for an AI is now falling into either (a) feeding from a data warehouse, lakehouse, lake, or data mesh or (b) taking masses of unstructured data from multiple silos and making sense of it.

We have done a lot of both, building our product, MyContentScout, as well as making use of Snowflake, AWS tools, and Azure tools - and they both present a different set of challenges.

With structured data, a lot of the work is around the more typical organising, ingesting, monitoring, aggregating, and tracing of the lineage of data. But there is also a new and developing set of AI tools that we can help you set up and use which allow on-the-fly analytics of the structured data. These can move the need for creating analytics and reports from SQL developers into everyday questions and answers that anyone can ask. We can help you build these bridges and implement these tools, which turn everyday speech into code “under the hood,” which in turn generates the just-in-time answers to the questions you ask and even creates dashboards on the fly.

With unstructured data such as photos, videos, documents, and social media feeds, we have done a lot of work in how to prepare this data for AI to successfully make sense of it. Each client’s needs are different but there are many similarities and almost all cases require some preparation. 

For example, by building a system that looks at documents from the last thirty years, we knew that the ingestion process for this data has to understand relevance, meaning, context, often language and business-specific terminology, and acronyms. It needed to understand relationships across documents and document types. Through MyContentScout, we have faced and overcome many of these issues for clients of wildly different sectors and with significantly different sets of raw data. We can help you in the preparation of this and guide you through the maze of vector databases and graph databases that help you get the best and most measurably reliable results.

We can also help in the deployment training and maintenance of data quality agents that can inspect and make recommendations on the quality of your data, which services both traditional services and other AI services further down the pipeline.

PILLAR 3: AI Governance 

The recently passed European AI Act is undoubtedly just the first piece of legislation that will impact all AI models and agents. Our in-house governance experts have read, understood, and thought through the implications of this act and can help guide you on what is required, advised, and suggested. We also apply their oversight to all the work that we do with any LLMs as well.

However, that is just the tip of the iceberg. We can help you put in place mechanisms and structures to manage the risks of AI, monitor the outputs (and the costs), understand and measure impact, and have lines of accountability that are matched to the improvements in efficiency that AI offers.

We can even go beyond that. How does GDPR factor into AI models and their training? How can you measure the carbon units consumed by your LLMs for ESG reporting or internal stakeholders? How can you convey the impact of AI to senior stakeholders? We have many years of experience helping build governance in many IT projects, and experts who have up-to-date knowledge of all these areas to help make sure that the efficiencies that get delivered are aligned with the values and outputs of your teams and organisations.

Conclusion 

We believe that by working through these three pillars, The Virtual Forge can offer a complete solution for both the technical implementation and the management of business risk when bringing Artificial Intelligence into your organisations.