Reversing the V: AI-Assisted Recovery of Requirements and Architecture from Legacy Systems

In textbooks, Model-Based Systems Engineering flows down the V-Model: idea → requirements → architecture → design → implementation. In industry, the greenfield case is the exception. Most engineers inherit decades-old systems with living firmware, CAD models, and schematics — but no requirements, no architecture, and no traceable design intent. When the "how" exists in full but the "what" has been lost, every change becomes a non-deterministic gamble.
This talk presents practical experience in reversing the V-Model with AI assistance. Starting from the delivered solution — source code, CAD data, and schematics — a test framework is generated first, anchored to the observable behaviour of the system as built. These executable tests then serve as ground truth for AI-assisted derivation of requirements and reconstruction of a coherent system architecture.
The approach is demonstrated live on an open-source embedded device, exercising both firmware and hardware against AI-generated tests on a reproducible bench. Attendees will see where AI genuinely accelerates the climb back up the V, where human judgement remains essential, and how the method reframes V&V and requirements engineering when specifications must be discovered rather than written.