PCB Copilot
Inspiration
As an electrical engineer, I ran into this problem firsthand while building a buck converter for my rocketry lab’s avionics team. What should have been a straightforward design task kept turning into hours of manual overhead: digging through datasheets, finding the right application circuits, sourcing passives, checking whether parts were actually in stock, and making sure every small component was included before ordering.
A lot of the delay was not “real engineering” in the creative sense. It was friction. We would miss small but critical parts like pull-down resistors, then have to revisit sourcing and update the schematic again. That created avoidable mistakes, wasted team time, and slowed down iteration.
That experience inspired PCB Copilot: a tool that reduces the gap between engineering intent and a buildable design. Instead of spending hours pathing through PDFs and distributor sites, engineers should be able to move from idea to validated starting point in seconds.
What it does
PCB Copilot is an AI-assisted hardware design workflow that helps transform a circuit concept and a datasheet into a practical starting design.
It focuses on:
- extracting useful design information from technical documentation
- helping calculate or infer supporting component choices
- sourcing parts that are actually purchasable
- visualizing the design so engineers can understand and iterate faster
The goal is simple: reduce the time spent on repetitive design setup and BOM acquisition, so engineers can spend more time actually building.
How we built it
We built PCB Copilot as a modern web application centered around speed, interactivity, and multimodal reasoning.
Core stack
- Next.js + React + Tailwind CSS for the frontend
- Server-Sent Events (SSE) for live pipeline streaming
- Vercel for deployment
- AWS Bedrock for model hosting and orchestration
- Pure SVG for wiring diagrams and lightweight visual modeling
- Auth0 for authentication and authorization
AI stack
- Google Gemini API for multimodal understanding across datasheets, schematics, and technical visuals
- grounded search workflows for validating engineering context
- structured design generation for sourcing, calculations, and ideation
Product architecture
The system is designed as a pipeline:
- ingest technical inputs such as datasheets and design requirements
- extract relevant engineering context from dense documentation
- generate or validate component-level design decisions
- connect those outputs to sourcing logic and visualization
- stream results live to the frontend for a more interactive experience
This made the product feel less like a chatbot and more like an engineering copilot.
What we learned
One of the biggest lessons from this project was that the real pain in electrical engineering is often not just the math or theory, but the workflow around it. Engineers already know how to design circuits. The frustrating part is the manual overhead required to turn that knowledge into a complete, sourceable design.
We also learned that usability matters as much as intelligence. A powerful model is not enough by itself. The interface has to make engineering decisions understandable, actionable, and fast. That pushed us to think beyond text outputs and toward a more visual, interactive system.
On the technical side, we learned how important it is to combine multimodal AI with a structured frontend pipeline. Raw model output is useful, but it becomes much more valuable when it is streamed, visualized, and tied to real design tasks.
Challenges we faced
The hardest challenge was translating messy real-world engineering workflows into something reliable and intuitive.
Datasheets are dense, inconsistent, and often difficult to parse in a structured way. Important equations, reference designs, and constraints can be buried across pages of tables, figures, and notes. Building a system that could extract the right information consistently was a major challenge.
Another challenge was sourcing. It is not enough to identify the “right” part in theory. In practice, engineers care whether the part is in stock, orderable, and realistic for procurement. That meant our system had to think beyond design logic and into real-world availability.
We also had to balance technical depth with simplicity. The product needed to feel useful to experienced engineers without becoming overwhelming for newer ones. Designing that balance was one of the most important product decisions we made.
Finally, streaming a multi-step AI workflow into a responsive frontend required careful system design. We wanted the experience to feel immediate and interactive, not like waiting on a black-box model response.
Why this matters
In hardware, even small inefficiencies compound. If a design loop takes hours when it should take seconds, teams iterate less, students get blocked earlier, and engineering time gets spent on procurement friction instead of product development.
PCB Copilot is our attempt to fix that. It is built from a real engineering pain point, shaped by firsthand experience, and designed to make hardware development more accessible, faster, and more intuitive.
In short: we are not just building an AI feature. We are building a better interface for electrical engineering.
Closing
PCB Copilot came from a simple observation: too much valuable engineering time is wasted on tasks that should already be automated. By combining multimodal AI, sourcing-aware workflows, and interactive visualization, we created a system that helps engineers move from concept to design-ready output far more efficiently.
That is the problem we experienced ourselves, and that is the problem we set out to solve.
Built With
- amazon-web-services
- and-design-generation-cloud-/-deployment:-vercel
- auth0
- bedrock
- html/css-frontend:-next.js
- javascript
- languages:-typescript
- react
- search
- svg
- tailwind-css-streaming-/-app-ux:-server-sent-events-(sse)-ai-/-apis:-google-gemini-api-for-multimodal-reasoning
- visualization:
Log in or sign up for Devpost to join the conversation.