PatSeek Patent Retrieval

Problem

Patent searches often fail when novelty, FTO, and technology research are treated as the same keyword task. Broad “object + generic function” queries produce noisy hits, weak evidence, and reports that are hard to audit.

How It Works

PatSeek routes work into four flows: novelty search, FTO, technology research, and direct lookup. Before calling APIs, it checks the technical object, dates, jurisdictions, document types, and confidentiality risk. During retrieval, it emphasizes IPC prechecks, term relationships, independent Boolean blocks, and hit-count gates. Semantic search is used only for candidate recall in the China library, submitted serially with checkpointed results. Evidence review aggregates patent families, records claim/paragraph locations, date roles, and confidence, and keeps conclusions scoped to databases, dates, and query terms.

Boundaries

It supports patent-document recall, candidate-pool construction, and report checking. It does not replace official registers, prosecution files, non-patent publications, or jurisdiction-specific legal-status verification; sensitive disclosures still require careful cache and API handling.