Enhanced Radar Is Building AI for Air Traffic Control and I Think They Might Actually Pull It Off

The Macro: Air Traffic Control Runs on Human Eyeballs and Cold War-Era Tech

Air traffic control is one of those systems that works well enough that most people never think about it. Planes take off, planes land, nobody crashes. The system is safe. But safe and efficient are different things, and safe and scalable are different things too. The FAA’s core radar and communication infrastructure dates back decades. Controllers still rely heavily on voice communication, manual handoffs, and pattern recognition skills that take years to develop. The US processes roughly 45,000 flights per day, and the number keeps climbing.

The modernization effort, NextGen, has been underway since the mid-2000s. It has cost billions. It is perpetually behind schedule. Anyone who has followed government IT procurement knows this pattern: the requirements are enormous, the vendors are entrenched, and the technology choices get locked in years before deployment.

Meanwhile, computer vision and natural language processing have gotten genuinely good at tasks that look a lot like what controllers do. Tracking moving objects on a screen. Parsing noisy audio for specific callsigns and instructions. Flagging conflicts before they develop. The gap between what AI can do in a lab and what it’s allowed to do in an operational ATC environment is wide, but it’s not a technical gap. It’s a regulatory and institutional one.

That’s the opening Enhanced Radar is trying to squeeze through.

The Micro: Pilots Building Tools for Controllers

Eric Button and Kristian Gaylord founded Enhanced Radar out of Y Combinator’s W25 batch. Eric is a former professional pilot with 2,200 hours of flight time and a G280 type rating. He previously co-founded Contrast, a fintech company that was acquired by PublicSq. Kristian comes from the computer vision side. He built embedded vision systems for satellites at Pilot AI (also acquired) and did high-frequency trading work in Rust before that. He studied statistics and computer science at Columbia. He’s also an FAA private pilot.

The fact that both founders have aviation backgrounds matters more than it might seem. ATC is a domain where credibility is earned slowly. Controllers and pilots speak a specific language, literally. The radio phraseology, the procedures, the failure modes are all deeply specialized. Building products for this world without understanding it from the inside is a fast way to build something that looks impressive in a demo and falls apart the first time a controller has to deal with a go-around during a thunderstorm.

Their product, Pattern, is positioned as an aviation operational intelligence platform. The company is also developing specialized training datasets for aviation AI, with what they describe as “aviation-grade quality standards.” They control the full pipeline, from data collection to in-house labeling and quality assurance. That’s a deliberate choice. Generic datasets built through industry-agnostic methods tend to miss the nuances that matter in aviation. A system trained on general speech data will struggle with the compressed, jargon-heavy exchanges between pilots and controllers.

They also run ATC.com, a live air traffic control radio service. That’s interesting as both a product and a data source. Live ATC audio is publicly available, but building a clean, structured dataset from it is non-trivial work.

The team is small, just two people, but they’re hiring aggressively across AI research, software engineering, aviation operations analysis, and iOS development. Garry Tan is listed as their primary YC partner, which signals that the YC leadership sees this as a significant bet.

The Verdict

I’m cautiously optimistic about Enhanced Radar, and I want to be specific about why the caution exists.

The technical challenge is real but solvable. Computer vision for radar tracking, NLP for controller-pilot communications, conflict detection algorithms. These are hard engineering problems, but they’re the kind of hard that gets easier every year as models improve and compute costs drop.

The harder problem is the market. Aviation is a regulated industry where the sales cycle is measured in years, the procurement process is Byzantine, and the cost of failure is measured in lives. The FAA is not going to rip out existing systems and replace them with a startup’s AI overnight. The path to revenue almost certainly runs through augmentation, not replacement. Tools that help controllers do their jobs better, not tools that replace controllers.

At 30 days, I’d want to know who their first customer is. Military? Private aviation? A specific airport authority? The beachhead matters enormously here.

At 60 days, I’d want to see how their datasets compare to what MITRE and other established aviation research organizations have built. Data quality is where this company lives or dies.

At 90 days, the question is whether they’ve found a procurement pathway that doesn’t require a five-year sales cycle. If they’re selling to the DoD or private sector ATC providers, the timeline compresses significantly compared to selling directly to the FAA.

The founding team has the right background. The problem is real. The timing, with AI capabilities maturing just as ATC infrastructure desperately needs modernization, is genuinely good. I think they have a shot. But this is a marathon, not a sprint, and the course is uphill.