Circuit Bench

Inspiration

I built CircuitBench because I wanted to enable AI agents to help out in scientific research. We can offload the researchers by letting AI agents handle the repetitive tasks like conducting iterations experiments and drafting their results. We can take this a step further by making agents decide what experiments to do and how, essentially making them an active participant in research.

What it does

Circuit Bench is a virtual electronics labaratory which expose WebMCP tools that let agents build circuits, run experiments, draft results and learn from the results to iterate and reach a goal.

How we built it

I built it with a Next.js frontend, FastAPI backend, ngspice for real circuit simulation, and WebMCP so agents can directly operate the lab.

Challenges we ran into

The hardest part was making sure the agent and the human interface always shared the same circuit state. Also the UI/UX took time to as the circuit canvas needed a clean and optimized interface.

Accomplishments that we're proud of

AI agent can independently experiment, learn from simulation results, and verify/improve its design. This loop can lead to very interesting findings.

What we learned

WebMCP has huge potential and it deserves to be explored more. Such a simple solution that can huge huge impact.

What's next for Circuit Bench

I'd like to support more components, richer experiments, larger circuits, and more advanced agent-driven engineering workflows. We could take this into other scientific domains too, not just electronics.

PS., I know the project name sounds AI-made but I named it myself, and I couldn't come up with something better (maybe we are starting to think like AI now)

Built With

• docker
• fastapi
• next.js
• ngspice
• python
• react
• restapi
• spice
• sqlite
• typescript
• webmcp