Overwatch AI Wants to Replace the Pre-Flight Information Scramble with a Single Natural-Language Query
There’s a specific kind of friction that anyone who has spent time close to airline operations knows well: the pre-departure information scramble. Before a flight moves, crews and ops staff are typically pulling from a patchwork of systems — aircraft manuals, company ops documentation, weather feeds, NOTAMs, airport procedures, ACARS messages — none of which talk to each other, and all of which demand context-switching that compounds under time pressure. It’s one of those problems that sounds mundane until you watch it slow something down that shouldn’t be slow.
Overwatch AI, a startup founded in Barcelona in July 2025 by former airline pilot Leo Kotil and technology entrepreneur Nikita Kaeshko, raised $1.5 million in pre-seed funding in May with a bet that natural language AI can fix exactly this. The round was backed by United Airlines Ventures, Baobab Ventures, Pegasus Innovation Lab, and Masia, alongside angel investors. The investor list is notable: Pegasus Airlines Innovation Lab became both a design partner and investor in the platform, with the airline contributing real-world operational context, extensive documentation, and direct crew feedback to tailor and scale Overwatch AI within a live airline environment.
What the Platform Actually Does
Overwatch AI consolidates fragmented aviation documentation, operational data, safety procedures, compliance information, and weather intelligence into a single AI-powered platform. The core interaction model is a natural-language query interface: teams can search and retrieve information using natural language queries, with all answers sourced from official documentation to support compliance and safety decisions. The practical framing from Kotil, a former pilot himself, is direct: after nearly a decade as a pilot, he described the experience as trawling through various apps and systems based on outdated software, trying to piece together the information needed to take action — spending more time on this than flying the plane.
The platform isn’t positioned as a standalone AI assistant but as a retrieval and synthesis layer over airline-specific documentation and live operational data. That’s a technically important distinction. General-purpose LLMs hallucinate because they generate from learned patterns; retrieval-augmented systems grounded in a specific document corpus are meaningfully more constrained, which matters a lot when the answer to a query might affect a go/no-go call or a MEL deferral. Overwatch has built a decision support system fusing a variety of sources together: company guidelines, aircraft manuals, and live APIs such as flight and runway data.
The RAG architecture feels like the right call for this use case — but whether it performs as intended depends heavily on the quality and consistency of each airline’s underlying documentation. That’s not a knock on the approach; it’s just the reality that documentation hygiene varies considerably across carriers, and a retrieval system is only as reliable as the corpus it’s pulling from. The risk isn’t architectural so much as it is operational, and it will play out differently airline by airline.
The platform is already operational across multiple airlines, handling over 30,000 flights monthly. Based on live usage to date, Overwatch AI said its platform saves airlines up to $4 million per year and yields a 6.6% increase in crew productivity.
The Design Partner Model — and What It Signals
The Pegasus relationship is worth pausing on. Having a carrier contribute its actual operational documentation and crew feedback in exchange for equity and early access is increasingly how the smarter airline AI startups are being built — it’s the difference between training on synthetic flight ops data and training on the real article. Pegasus was among the first airlines to test Overwatch’s technology. That kind of embedded design partnership compresses the feedback loop considerably and gives the startup a defensibility argument that’s harder to replicate than the model itself.
From time spent running global delivery teams for airline software, I’d say the honest trade-off here is this: a product built this closely with one carrier will almost inevitably reflect that carrier’s workflows and documentation structure more than anyone else’s. That’s not necessarily a problem — it gives Overwatch a well-tested baseline and a reference customer with real operational credibility — but the degree to which it generalizes to other airlines will come down to how deliberately the team designs for configurability as they scale. It’s an execution question as much as a product one.
I recently polled aviation industry professionals on where AI in the cockpit should go first. The results split almost three ways: 40% said maintenance and documents, 20% said pilot decision support, and 40% said AI should be kept out of the cockpit entirely. That last number is striking. My read is that it reflects a genuine feeling of uncertainty and distrust around AI in high-stakes environments — which is understandable, and not irrational. A documentation-retrieval tool like Overwatch, though, feels categorically different from the autonomous or advisory AI applications those respondents are probably most wary of. It’s not making decisions; it’s surfacing the information that crews are already required to consult, just faster and through a single interface. That distinction might make it a reasonable entry point — a way for AI to earn credibility in the flight environment incrementally rather than all at once. Over time, I’d expect skepticism to soften as these systems accumulate a track record, but that trust has to be built carefully and on the basis of demonstrated reliability, not just marketing claims.
Everyone wants to deploy AI solutions, but everyone is scrambling for the data points needed to evaluate these systems. That gap between the pitch and the proof point is real, and the 30,000-flight-per-month figure gives Overwatch something concrete to offer prospective airline customers. Whether the productivity and cost figures hold across fleets with different documentation architectures and workflow cultures is the next question worth watching.