Inside CesiumAstro's AI Acceleration: Joe Ellis on Scaling Intelligence Across Space Systems

Machine learning scientist Joe Ellis joined CesiumAstro following its acquisition of Vidrovr, an artificial intelligence (AI) company specializing in real-time multimodal signal analysis. In this conversation with Aimée Ahiers, CesiumAstro's Director of Marketing, Joe shares his path from MITRE and Columbia to building AI-native space systems, and how intelligence embedded across satellites, payloads, and ground networks will reshape how space infrastructure is designed, operated, and scaled.
‍

Aimée Ahiers (AA): Joe, it’s great to have you with us at CesiumAstro. Tell me about your background.

Joe Ellis (JE): Thanks, Aimée. I am excited to be here! I started my career at MITRE, where I worked as a communications engineer supporting the Joint Tactical Radio Systems program. I had always been fascinated with video and how to help computers understand it, so I left MITRE to pursue a PhD at Columbia University. While I was at Columbia, my colleagues and I invented and patented a system and machine learning (ML) algorithms for near-real-time understanding and searching through massive amounts of video and other multimodal signal data. I worked at IBM Research and Google but eventually decided with my co-founder to spin out the technology we developed at Columbia to start Vidrovr.
‍

AA: Can you tell us why you are excited about what we’re building here?

JE: Oh my gosh, where to start? From the first time I met Shey [Sabripour, Founder and CEO of CesiumAstro], I was inspired by his vision to make “information accessible to all.” Our vision at Vidrovr was to “make the world’s video information understandable.” I was struck by how naturally those ideas align. Tremendous amounts of data now and in the future will flow through CesiumAstro’s communications systems, whether from Element satellites, phased array payloads, or ground stations. That data is ripe for machine learning and AI analysis, enabling more efficient transmission, autonomous satellite operations, and new applications for the space industry.

I’m impressed by the caliber of individuals here. CesiumAstro has some of the world’s leading experts in space infrastructure and technology development, including Cody Vaudrin and Thomas Magesacher, and collaborating directly with them has been a tremendous opportunity for our group. I’m energized by the opportunity to apply ML and AI to help build what comes next.

‍AA: Where do you think the space industry operates with a "legacy mindset?" What needs to change?

JE: I think there are three areas the space industry should embrace.

First, I am excited about new compute architectures for on-orbit computation. Traditionally, the space industry relied on low-power FPGA architectures for onboard compute, but with the advent of new GPU, TPU, and NPU processing chips, we can offload compute into these new computing architectures, enabling novel applications with enhanced edge processing capabilities.

Second, flying a satellite is a manual process. I think industry should lean into investigating ways launching, controlling, and monitoring satellites can be automated to lighten satellite operators' cognitive load.

Third, I think the space industry should embrace multimodality in algorithm development. Satellites are equipped with expensive visual, RF, and EM sensors. Each of the data generated from these sensors is typically analyzed in a silo. By fusing the information across each, we unlock new applications or improve performance in existing applications.
‍

AA: Some say we're entering the golden age of space technology development. How does the proliferation of assets in space affect the way you think about AI application?

JE: AI and ML are the fundamental tool to make sense of all the assets in space moving forward. We are seeing an exponential increase in the number of satellites launched, and this expansion will only accelerate. Each satellite is equipped with more sensors, causing an explosion in the amount of data created. We don’t have enough skilled operators to fly, position, and command our satellite and communication systems effectively.

AI can enable a world where every satellite or comms system self-operates without a fully staffed mission operations center and network operations center. Given the increase in launches, these types of automated decisions are the only way we can control the massive proliferation of new assets over the next decade.
‍

AA: You mentioned edge processing. How are you approaching that?

JE: There are a ton of new compute systems available allowing us to perform optimized inference of our machine learning algorithms on-device. We’re diving deep into the world’s most complex, high-powered system-on-chip devices to see how we can optimize algorithms never before implemented in space. This is easily one of the most exciting parts of joining CesiumAstro. Stay tuned for more.
‍

AA: What new capabilities will customers see as CesiumAstro embeds AI/ML into our products?

JE: It's a great question. CesiumAstro is unique in that we have products up and down the satellite communications spectrum. I expect to see the AI/ML team's capabilities impact every portion of our product line. Our algorithms will improve efficiency in our phased array systems, as well as the ability to operate reliably in adversarial RF environments.
‍

AA: Looking ahead, how are you planning to scale your team? What do you think will be the most challenging part?

JE: The AI team is aggressively hiring. CesiumAstro is one of the best places to work to solve the most important problems in space communications today. Our team is a mix of hackers, PhD scientists, mathematicians, and engineers. Everyone is different, with a unique background and skill set, making working together fun. The hardest part about building ML algorithms for distributed satellite communications is making the complex pieces and specialties work together. You need expertise in designing machine learning algorithms and implementing them efficiently on low-level hardware, analyzing many disparate sensor types, and orchestrating tens of disparate systems in one codebase. It’s a daunting task, but one of the most intellectually interesting problems I have encountered.
‍

AA: What should the public understand about the impact of AI in this industry and its role in everyday life?

JE: Building hardware and software to operate in space is hard. It takes tremendous ingenuity, effort, and, to be honest, funding. If you can build reliable systems, the advantages are extraordinary. Here in Austin, where there is exceptional cell coverage, Waymos are everywhere. This type of autonomous driving is impossible where cell coverage lapses. In the future, the industries changing the world today will rely on a robust satellite communications system. It’s exciting to be working on the bleeding edge of this society-transforming technology.
‍

AA: What will success in space look like 5, 10, or 20 years?

JE: Success to me looks like CesiumAstro operating large-scale space communications systems that are autonomous, resilient, and adaptive. In 5 years, I expect AI to be deeply embedded across our product lines, supporting autonomous operations and decisionmaking. In 10 years, I believe CesiumAstro will operate intelligent space networks that dynamically optimize in real time. And in 20 years, AI-enabled space infrastructure will underpin entire industries on Earth, from communications to transportation to global logistics.
‍

AA: Thank you for spending time with me today, Joe.

JE: Thank you for having me. I’m very excited to be here.
‍

AA: Let’s catch up in a year and see how far things have progressed and hear your industry insights.

JE: Deal.