Collimate Space, a new Silicon Valley startup, is tackling a specific problem for satellite operators. The company offers a tool for predicting the success of satellite downlinks, factoring in space weather, terrestrial weather, and the location and profile of ground-based antennas.
This is valuable because satellite operators can confidently tell their customers, “I can get you the data with 90-percent confidence,” explained Guillermo Jenaro, Collimate co-founder and CEO. “It gives you more precision in your prediction. This will empower our customers to say yes to customers they could not assist before.”
Beyond predicting success rates, satellite operators can use Collimate's tool to "reverse-engineer" their operations to increase the odds of success, said Jenaro. "Maybe it’s going to require that you start sending data earlier or have two ground stations in your scheduling system."
Jenaro, who spent nearly two decades with Airbus across Spain, France, Germany and the United States, founded Collimate with Anthony Xiao, former Slingshot Aerospace engineering vice president. They met while both were at Acubed, Airbus' Silicon Valley innovation center.
Collimate aims to develop a range of atomic services for satellite operators, with the scheduling product being just the first step. “Whenever we design anything, we make sure that you can plug it in with any interface you use,” Jenaro said. “We have a unique approach to this informed by having been at the customer side operating large, complex aerospace products and understanding the ecosystem.”
The company leverages deep learning "to take what has historically happened, fuse it across lots of data and anticipate future events," according to Jenaro. "This can apply to scheduled communication, or avoiding jamming and space weather interference."
Collimate plans to use artificial intelligence to create optimized communications planning for space operators in the future, Jenaro shared. "In the future, there will be several layers of AI throughout the tech stack and chat interfaces might be powered by" large language models.