My team and I built an AI-powered system that converts breadboard prototypes into manufacturable PCB designs. The goal of our project is to help makers and engineers bridge the gap between rapid prototyping and production by automating much of the PCB design process.

Our system combines three sources of information: a CubeMX configuration file that specifies the microcontroller and pin assignments, a photo of the breadboard prototype, and a component list. Using the DigiKey API, we retrieve datasheets and product images for each component, which provide additional context for our computer vision pipeline. We then use the Gemini API to reason over all of this information, infer the circuit structure, and generate a schematic and PCB layout.

A second Gemini agent is responsible for abstracting away prototyping hardware and extracting only the essential components needed for manufacturing. For example, if the prototype uses an STM32 development board, the agent replaces it with the underlying STM32 microcontroller and its required supporting circuitry. Similarly, if the prototype uses a microphone breakout board, the agent identifies the actual microphone component and removes the breakout hardware. This ensures the final design reflects a real manufacturable circuit rather than a prototyping setup.

The final output of our system is a KiCad PCB file that can be opened, edited, and manufactured using the KiCad ecosystem. By combining computer vision, large language models, and hardware design tools, our project streamlines the process of turning a messy prototype into a clean, manufacturable PCB.

Next steps for the project include automating trace routing while respecting electrical and layout constraints, improving the reliability of schematic generation, and extending the system to automatically insert necessary supporting components such as decoupling capacitors and other small passive components required for stable real-world operation. These improvements aim to make the system fully end-to-end and production-ready.