Who's Kevin?

You’ve probably heard promising words from both high and low about so-called LLMs (Large Language Models) by now, but the question we have asked ourselves is: what can they actually do for us in Solution Seeker?

The short answer? Quite a lot - both internally and externally.

Internally, we maintain a data model designed to mirror our clients’ production systems: nodes, sensors, connections - the whole shebang. Keeping this model up to date is largely manual, time-consuming, and frankly unnecessary given today’s technology. The same goes for data analyses. Setting them up often becomes an exercise in overhead, only to discover mid-way that new research questions and hypotheses keep popping up, requiring additional time-consuming alterations. What should have been a quick, ad-hoc analysis, becomes an effort-draining task indeed. But such on-demand analyses aren't limited to us, they're just as valuable to our clients - which brings us to the next topic: external value. Externally, our clients increasingly want transparency into our models, technology, and algorithms. They want to understand how things work, how results are governed, and what data underpins our methodologies. Documentation, however, is notoriously difficult to keep current and relevant for all user requests - especially when it’s static in format. If only it could be dynamic...

The common denominator? Text. All of these use cases revolve around text - sometimes as the primary data source and other times as contextual input layered on top of conventional quantitative data. With LLMs, suddenly both we and our clients can exploit the powerful combination of textual and numerical data through simple, natural queries.

For us, this is very much a greenfield initiative - and one we’re taking seriously. As American Authors put it: go big or go home. To signal our commitment, we’re investing heavily. In the works, is a PhD (!) on the topic by our very own senior data scientist, Kristoffer Nesland, which is in collaboration with UiS (Universitetet i Stavanger), the Norwegian Research Council, and Solution Seeker.

In January, we’re also welcoming three interns for a five-week, LLM-focused winternship. In true Occam’s Razor fashion, they’ll start with a narrow, well-defined scope: well test transparency and ad-hoc analyses of well test data. They’ll explore how Kevin can answer questions like:

• How much has well production declined over the past year
• What is the percentage increase in water cut and gas–oil ratio over the last six months?
• How has artificial gas lift affected the well in recent weeks?
• What is the average uncertainty of the test separator?

The data needed for these analyses already exists in our Well Test Application. And so, the real magic isn’t what Kevin outputs - it’s that the user doesn’t have to lift a finger beyond asking the question, and gets an answer almost instantly.

Beyond that, it’s very much all hands on deck. Both the Oslo and Brazil offices are involved in this initiative, and rumour has it our UX designers are bringing some much-needed front-end wisdom to the table. After all, even the smartest LLM is useless if no one wants to use it. Back-end brilliance means little without an equally thoughtful user interface.

We’re eager to dive deep into this space - and to blow your minds along the way! Stay tuned. Exciting things are coming 🚀