Local USPTO corpus pipelines for search and training prep

Building a locally owned patent corpus pipeline with streaming parsers and normalized exports.

Research question

Can weekly USPTO XML dumps be stream-parsed into locally owned SQLite/JSONL artifacts suitable for search and downstream ML prep without loading archives into memory?

Hypothesis

Streaming normalization enables a personally owned patent corpus at multi-terabyte scale with inspectable provenance.

Methodology

PatentPulse downloads official dumps, stream-parses, writes normalized outputs. Scale claims follow repository documentation only.

Results

Pipeline approach is implemented in-repo. Treat corpus size and training outcomes as repository-documented facts, not site marketing.

Limitations

Requires large local storage; legal/use constraints remain the operator's responsibility.