Manually translating unstructured aerospace data—like PDFs, documentation, and lookup tables—into verified simulation models is traditionally a slow, error-prone process that takes days of engineering effort.
In this video, we showcase a complete, closed-loop demonstration of the Dyad Agent recreating the historic NASA HL-20 lifting-body spaceplane. Instead of hand-coding from scratch, the AI agent is given the exact raw materials a human systems engineer would receive: a basic README scope, an unstructured PDF specification containing aerodynamic coefficients and vehicle geometry, and a raw DML data file with lookup tables.
Watch a full reasoning trace as the agent independently parses the documentation, maps complex control mixers, writes native model representations, and executes inline Julia code to continuously verify and refine its own work. Finally, see the model undergo automated validation by running a pitch-pulse simulation against NASA's original reference data to prove real-world accuracy—all accomplished within a single, continuous workflow.
