Language-Model Agents for Photonic Integrated Circuit Design — Thesis Data

The thesis measures which agentic mechanisms — knowledge-graph grounding, formal topology gating, and self-critique — improve end-to-end photonic integrated circuit design on three language models, under one blind three-axis human review.

Deposits include the benchmark suite with expert-audited golden references, the full evaluation run corpora with blind-review grades and generated schematics, agent execution traces, and a complete knowledge-graph export with verified reconstruction tooling. Accompanying open-source code: github.com/JPPhotonics/PhIDO-Agentic