The Center for Data Innovation recently spoke with Jayesh Gupta, CEO of Silurian, a Seattle-based company developing an AI model that offers predictive intelligence for weather-driven risk and operations. Gupta explained how Silurian’s model integrates forecasts with hyper-local and operational data to help energy providers and emergency management services anticipate and respond to risks such as flooding and icing.
David Kertai: What does Silurian offer?
Jayesh Gupta: Silurian provides a global and regional weather forecasting model, accessible through application programming interfaces, that delivers higher accuracy than comparable models. Our core technology is called the Generative Forecasting Transformer (GFT). Beyond general forecasts, we offer specialized predictions that strengthen a host of different types of preparedness. By integrating hyper-local data into GFT, we tailor forecasts to each customer’s assets and operations, enabling decisions that are both timely and highly specific.
Kertai: How do your forecasts differ from others?
Gupta: Our forecasts differ in three key ways: cost, speed, and customization. National agencies depend on massive, centralized supercomputers to generate broad, one-size-fits-all forecasts. By contrast, GFT produces results faster and at far lower computational cost using only a GPU. We combine global context with hyper-local data, delivering personalized forecasts down to the level of a feeder line, substation, or farm field, something national centers aren’t built to provide. We see the future of weather intelligence as personalized and integrated across multiple scales, not delivered as distant, uniform products.
Kertai: How does GFT work?
Gupta: Think of GFT as “a language model for the atmosphere.” Just as ChatGPT learns the grammar of language, GFT learns the grammar of weather by training on vast amounts of environmental data. It predicts one spatiotemporal state—the condition of the atmosphere at a specific place and time—after another, generating a rolling forecast that flexibly incorporates many types of input data. This efficiency lets us continuously update forecasts and fine-tune them to reflect the microclimates and risks most relevant to each customer.
Kertai: What has been one of the most impactful use cases?
Gupta: Forecasting transmission-line icing risk has been especially impactful. By combining historical icing observations with GFT, we can predict the probability of icing on a specific transmission line a day in advance. That capability lets utilities pre-position de-icing crews before storms hit, instead of scrambling afterward. This shift from reactive to proactive response is transformative. While most of our current customers work in energy and emergency management, we are also exploring applications in transportation and logistics, where weather-driven disruptions can impose enormous economic costs.
Kertai: What future capabilities excite you most?
Gupta: We are most excited about expanding the data streams that integrate with GFT. We aim to move beyond traditional weather data to include environmental, infrastructure, and operational inputs. Our long-term vision is to deliver not just forecasts, but end-to-end decision optimization, all powered by the same model. By building a unified platform that scales across industries, we can help organizations anticipate and manage risk more effectively, reducing inefficiencies and costs.
