Patent Mining Expert Agent

Problem

Before filing, engineers often manually search patents, compare prior art, extract invention points, and assemble scattered notes into a disclosure. This process can miss close references, frame innovation as product features rather than technical solutions, lack comparative data, and produce inconsistent documentation.

How It Works

The skill runs a structured workflow:
- Search & analysis: extract Chinese/English keywords and IPC/CPC classes, query CNIPA, Google Patents, USPTO, Espacenet, and similar databases, then build a prior-art list and gap analysis.
- Innovation mining: apply TRIZ to compare defects, technical contradictions, and solutions, producing “problem-solution-effect” invention points.
- Evidence & visualization: use matplotlib, seaborn, plotly, and related libraries to generate performance comparisons, radar charts, and process diagrams.
- Disclosure drafting: structure background, technical solution, claims, and effects, then archive files and generate a README.

Boundaries

It is useful for organizing patentable points from an existing technical proposal, but it does not replace patent counsel for infringement conclusions or legal advice. Search quality depends on public database coverage and keyword/class strategy; broad topics or thin technical details need added experimental data, system boundaries, and comparable references before final drafting.