The start-up Reshape Systems is developing an artificial intelligence tool for risk analysis, particularly in the aerospace and nuclear industries. CEO and co-founder Andrea Apollonio explains the company’s approach.
Reshape Systems aims to make risk identification more efficient and reliable in high-tech sectors such as aerospace, nuclear energy, and industrial machinery. To this end, the company has developed an AI copilot capable of analyzing every component from the design stage onward, helping to mitigate potential risks and reduce their impact on costs and time-to-market. This innovative approach earned the start-up the Grand Prize at the 2025 edition of Tech4Trust, a Swiss accelerator that supports start-ups active in the field of digital trust, a market projected to reach nearly 1.2 trillion dollars by 2031, according to Kings Research.
How did Reshape Systems come about?
Andrea Apollonio: My co-founder, Thomas Cartier-Michaud, and I spent many years working on safety-critical systems for CERN’s particle accelerators, the European Organization for Nuclear Research in Geneva. This fascinating and unique environment made it clear just how challenging risk analysis can be for highly complex systems. The same issues arise in cars, smartphones, or drones. The pressure to shorten development cycles and accelerate time-to-market makes the task even harder.
What exactly does your "risk-analysis copilot" do?
Apollonio: It’s a tool that enables risk prevention right from the earliest design phase of a project. Risk analyzes are often carried out late in the development cycle, when every problem costs ten times more to fix than it would have if addressed earlier. Our copilot ideally intervenes from the initial design stage. It begins by automatically processing large volumes of technical documentation (text, diagrams, and plans), then breaks the system down into subsystems, identifying components, risks, and mitigation measures.
At each stage, the user validates the suggestions. It is a collaborative process in which the AI remains fully transparent. We show its reasoning, assign confidence levels to its assessments, and allow engineers to add their own expertise. Our approach is built on AI that effectively assists, rather than replaces, experts. In our pilot project at CERN, we achieved more than an 80% reduction in engineering costs, while also improving long-term knowledge management for the experts involved through the centralization of all risk-related information within the AI models.
How is AI changing the approach to risk analysis?
Apollonio: Traditional analysis involves manually identifying potential system failures based on technical documentation. Generative AI is therefore a powerful tool for processing multimodal data and producing advanced analyses. Our solution also delivers "explainable" AI results, ensuring that outcomes are both transparent and verifiable. Generative AI can perform many complex tasks, but it must be carefully managed to produce dependable results. We have therefore developed a suite of tools to control its operation and strengthen confidence in its conclusions.
What are the main technical challenges?
Apollonio: The first is to make generative AI reproducible and reliable. The second is to convince a sector that is traditionally conservative. The methods of risk analysis used today are almost identical to those applied during the earliest space missions. Our challenge is to introduce AI into these sensitive environments without compromising existing standards, and in fact to enhance them.
Which sectors are you targeting as a priority?
Apollonio: Our approach is relatively agnostic, as we automate a methodology rather than a specific technical field. Our first users come from the nuclear and heavy-industry sectors, but we are also working on projects involving industrial machinery and drones in Switzerland. We are keen to support green technologies and are developing several partnerships in that area.
What are the next steps for the company?
Apollonio: In the months ahead, we plan to consolidate our partnerships with our first clients, particularly in the nuclear sector, and to launch our first certified product on the market. Our ambition is to evolve from the traditional "Software-as-a-Service" model towards "Service-as-a-Software"; in other words, a copilot capable of automatically providing a service under user supervision. Finally, we will continue to promote our core belief that trust is the key to AI adoption.
