AI Computing &System Architecture


Agentic AI for Circuit Design Screening
Powered by SLM cross-referencing to safeguard circuit design reliability
Circuit Detective is a self-developed, on-premises LLM-powered microservice that delivers an end-to-end pipeline for automated circuit design verification. Once a user uploads a KiCAD or Cadence netlist, or a photo of a card-puzzle board, the system automatically constructs a Circuit Topology Graph (CTG), retrieves component datasheets via RAG (Retrieval-Augmented Generation), and applies a large language model to perform design-rule checks, streaming the results back in real time. At its core, the system runs a dual-track architecture of "LLM reasoning + deterministic rules": tasks requiring semantic understanding of circuit layouts, or judgments that cannot be reduced to exhaustive rules, are handled by an autonomous AI agent, which in turn invokes specific functions to enforce deterministic logic checks based on the layout at hand — the two tracks complement each other to balance flexibility with reliability. The system also renders an automatic circuit-schematic visualization, marking flagged issues directly on the diagram so problems can be pinpointed at a glance. The whole system runs independently of any network connection, deployable on on-site hardware such as Jetson Nano and RTX 5090 units, enabling rapid rollout on production lines and reducing manual inspection costs.
