LLM Agent for Analog Circuit Design with SPICE in the Loop

Can a small language model design analog filters and amplifiers when it works as a tool-using agent with the SPICE simulator in the loop? This project builds a tree-of-thought design agent on Google's Agent Development Kit: the model proposes component values with the help of a closed-form design_equations tool, ngspice 47 simulates every candidate, and a deterministic controller keeps the best designs and feeds the measured response back. It is compared with the textbook closed-form designer, Bayesian optimisation (Optuna), a single-shot LLM and a no-feedback ablation on an own seeded benchmark of 35 specs over 7 topologies. qwen2.5:3b was served by Ollama on a second laptop (RTX 2050 4 GB, 32 GB RAM) over Wi-Fi; the code ran on a laptop with RTX 2050 4 GB (Ryzen 7, 8 GB RAM).

NOTE: Teaching / research demo: designs come from a small language model and an idealised op-amp macromodel; check every circuit before building it.

Category: Department of Computer Science and Engineering / Agent Development

Agent Development Project AG09 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.