ÿ�մ���

A generative artificial intelligence (AI) platform that automatically produces detailed debriefs of each training flight has been launched by U.S. startup Navi AI, which has emerged from stealth having raised $6 million in funding from backers including United Airlines Ventures.

Using a small orange box installed in the aircraft, Navi’s system ingests cockpit audio, aircraft data, traffic and weather information and transmits it to the ground, where a domain-specific large language model analyzes intent, behavior and performance, and aligns its output with the flight school’s training syllabus.

Navi founder and CEO Nikola Kostic compares the process to how the National Transportation Safety Board (NTSB) investigates aircraft accidents or airline flight operations quality assurance (FOQA) programs analyze flight data from onboard recorders to detect safety trends.

“[The platform] figures out who’s flying the airplane, when was the transfer of control, which maneuvers are being demonstrated, which are being flown by the student, and figures out where the plane is. All of this is traditional FOQA analysis and is not AI,” he says.

“But when it comes to meshing all of that together to get the context, that’s where the AI comes in. Just like the NTSB does when there is a crash ... we do the exact same thing, but we do it with AI that is getting scarily good,” Kostic says.

Founded in 2024, Navi partnered with Embry-Riddle Aeronautical University, with its expertise in flight training and syllabus design, to help integrate AI-powered debriefing into structured pilot training courses. An equity stakeholder, Embry-Riddle is deploying Navi’s system into its training programs.

Navi launched its first commercial deployment in September 2024 with U.S. Part 141 flight school Sling Pilot Academy, where the AI debriefing platform has logged more than 55,000 flight hours annually supporting students and instructors from ground school through post-flight review.

The platform is also in use or under evaluation by the University of North Dakota, Purdue University, Delta State University and the U.S. Air Force Test Pilot School at Edwards AFB, California, where Navi is working under a $1.27 million contact to adapt the platform to the school’s Northrop T-38 fleet.

At a flight school, the platform goes through a calibration phase that takes two weeks and about 100 flights, where the instructors review everything that the AI outputs. “When we reach 95% accuracy on everything, then we go ahead and deploy. But we are constantly improving and adding things,’ he says.

“We ingest all their SOPs [standard operating procedures] as the source of truth. But the problem we’ve found is there is a lot of tribal knowledge in aviation and the SOPs don’t have everything. The flight schools find out through the calibration process that they didn’t specify something in their SOPs, so we end up improving them through that process,” Kostic says.

“We’re constantly validating [the model] with a study Embry-Riddle designed for us and we’re tracking 95-96% accuracy in terms of alignment with what the CFI [certified flight instructor] says. I suspect we’ll be at 98% soon and that last 2% might never get sorted because even a human ingesting all this data will never be able to figure that out,” he says.

“But that’s way more than useful for deployment to help students and instructors and help flight schools to zoom out and see how is their fleet performing? How are the pilots doing at this stage? What are they struggling with? What happened after we made an SOP change to reduce the risk?”

While the main product is an “FOQA on steroids” for flight schools, it is also a benefit for students and instructors. “When people think about AI and aviation they think you’re putting something on the plane that’s dangerous, that’s going to distract the pilot. We’re doing the opposite,” he says.

“We’re reducing workload because they don’t have to take notes and their debriefs can be a lot better. And all of this is saved for the student to review at home at night.” Students can use an AI assistant, grounded in the SOPs and FAA regulations, to ask questions and find tutorials tied to their performance.

To prevent the AI generating an output that could endanger the student, Navi incorporates an extra step. “AI is always geared to give you a response, even when it doesn’t know. So, we let it give the wrong response. But then we run it again and say did this actually happen and is this in the SOPs?” he says.

“We also have a bounty out that we’ll pay $100 to anybody that can make it make a mistake on the regulations. Nobody’s claimed it yet. And then we have a step where the flight instructor reviews the debrief first and forwards it to their student. So, we have a human in the loop at all times.”

Based in San Francisco, Navi has raised $6 million to date. In addition to United, the latest round included funding from Bravo Victor Venture Capital, New Vista Capital, Raptor Group and I2BF Global Ventures. The funding will be used to grow the company and expand its customer base.

Navi’s platform is available as a subscription service. “The hope is that, by the end of the year, we’re covering a decent chunk of the largest flight schools in the U.S.,” Kostic says. The startup has established a partnership to integrate the platform into Garmin’s avionics ecosystem.

Longer term, beyond training, Navi plans to bring its AI platform to commercial aviation, applying its real-time analysis and performance intelligence to airline operations to improve safety. “The idea was always to have this thing in as many cockpits as possible,” Kostic says.