I used to be the lead designer for a stellarator fusion startup. Zoo is the tool I wish existed - here's why. Modern design and simulations tools are tedious and segregated, and going from something that looks good in simulation to something manufacturable is itself an art of engineering experience. Design and simulations tools have become more feature-rich in recent years but haven't fundamentally changed all that much - its endless point-and-click operations. ML has accelerated software development by 10-100x but mechanical design hasn't kept up. There's two main methods by which hardware can become as fluid and quickly iterable as software: - Accelerating the design and simulation tools - Accelerating modular, flexible manufacturing methods Here's the simple reason I'm proud to be an investor in @zoodotdev - it has the potential to hit both of these directly on the head. There is no separation from design to manufacturing - this is why Bill Rich, who led Lockheed Martin's Skunkworks said engineers need to spend time with machinists - what might look good in a design might be impossible to manufacture, and small design tweaks can save a huge number of steps in tooling setup and machining methods. Similarly, there's no separating design from simulations. In modern contexts these are in constant feedback to hit performance specs across thermal, mechanical, and electromagnetic domains using both FEA and CFD simulation tools. At the heart of every design tools lies the Geometry Kernel, an often-under appreciated code behemoth that calculates the countless geometrical edge cases that enable a designer to put their ideas into mathematically defined geometries. Yet the majority of design tools today haven't fundamentally changed in decades - its a point-and-click interface where a scripting back-end may be available, but is often far-separated from the users familiar actions. What Zoo is fundamentally doing is several things: - Creating the first new professional geometry kernel in decades from the ground up intended to be accelerated by ML automation in generating mathematically defined geometries aligned with the users intention - Placing scripting and UI design methods on equal footing, so every UI design decision generates code automatically, natively training designers in how to think through a design programmatically. And in the near future: - Integrating simulations methods from FEA and CFD into the design process, so there is no clumsy translation back-and-forth with static STEP files between Solidworks / etc into COMSOL or ANSYS, where simulation-informed design changes require tediously re-assigning boundary conditions. Instead, a design can be simulated and optimized within the gradient space of its performance metrics automatically. And further out than that? - Piping the whole thing into manufacturing methods. Capture the designers intent, simulate to performance specs, and have contextual awareness of how something can be made, all in one process thats accelerated by ML. Everything is downstream of the design process that seeks to achieve a performance metric, bounded by what can be manufactured. Starting with the design, then moving into simulations, and then manufacturing is how you go from "Point and click tedious CAD" to "Jarvis, design a new repulsor jet" Today the text-to-CAD may just be simple gears and components, but soon larger complicated assemblies. I wouldn't bet against the compounding returns on user-generated data in accelerating the performance of ML-enhanced design tools.