Kevin Delivers
Five weeks ago, we introduced you to Kevin - our clever LLM, not the Home Alone kid. We laid out an ambitious vision: a production engineer assistant that could answer real questions about well performance, using nothing but natural language. We welcomed three interns, rolled up our sleeves, and got to work.
- Author
- Jens Nikolai Alfsen
- Publish date
- · 2 min read
From vision to product
In December, we listed a set of concrete questions Kevin should be able to answer: production decline, water cut trends, gas-oil ratios, gas lift effects, separator uncertainty. These weren't hypothetical. They were the exact questions our clients' production engineers ask on a daily basis, and the exact questions that used to require manual configuration, bespoke analysis setups, and a good deal of patience.
Today, Kevin handles all of them. Ask him: "How much has well production declined over the past year?" and he'll fetch the relevant sensor data, plot it, and give you a concise summary. No configuration. No overhead.
How much has well production declined over the past year?
He'll fetch the relevant sensor data, plot it, and give you a concise summary. No configuration. No overhead.
What Kevin actually does
Under the hood, Kevin is a conversational agent with deep access to our data platform. He understands the hierarchy of a production system - platforms, components, sensors, tags - and uses that understanding to navigate vast amounts of real-time and historical data on behalf of the user.
But Kevin doesn't just retrieve data. He presents it. Ask about well test results, and you get structured cards with oil, water, and gas rates. Ask for a trend, and an interactive time-series plot slides into view. Ask for a comparison, and a clean table appears - ready to copy into your report. The interface is designed so that the data speaks for itself, while Kevin provides the narrative.
The people behind
None of this happens without the right people. Our three interns dove headfirst into a complex, production-grade codebase and delivered features that are shipping to clients. In five weeks. From agent tooling and data integrations to evaluation frameworks and front-end polish - their fingerprints are all over this product.
The winternship model worked exactly as intended: a narrow, well-defined scope that still left room for creativity and ownership. The result speaks for itself.
Kevin is live and in the hands of real users. But this is just the beginning. The roadmap ahead includes broader data access, deeper analytical capabilities, and tighter integration across our ProductionCompass AI suite.
Our interns: Aimee, Oskar and Erik