PatClaim.ai (forthcoming AI-related & AI-based work with patent claims)

• Including using text of patent claims as proxies for patent spec as a whole
• Patent mining based on patent claims (the legally-operative “business end” of patents often overlooked in traditional research methods).
• Patent searching based on semantic similarity of claim language, rather than solely on keywords, classifications, citations, etc.
• Custom patent-claim LLM which, given patent claim text as input, outputs predictions e.g. likely grant vs. non-grant, likely litigated vs. non-litigated
• Using large language models (LLMs) and neural networks trained specifically on patent data to generate sentence embeddings (mathematical representations of text) for claim comparisons.
• Patent “landscaping” using patent claim sentence embeddings, and SentenceTransformer models (like PatentSBERTa), to map claims into a vector space, potentially identifying “holes” in the space (un-claimed technological areas).
• Uncovering prior art using patent-claim semantic similarity (claims are a limited form of prior art, but use claims as proxies for earlier disclosures, where the earlier disclosure is already represented in claim language that might better match the current claim at issue)
• Applying machine learning to patent-portfolio ranking, based on the claims (e.g. possible case study of patents of QQQS holdings; compare QQQS to QQQ — does diff QQQS minus QQQ in some way reflect value of patents?)
• Developing tools to query PTO PatentsView API and Google Patents with SQL (BigQuery).
• Examining patent-related AI, such as PQAI (AT&T, GreyB; API)
• Sessions with ChatGPT (and Anthropic’s Claude) developing Python code to work with patent claims and patent-litigation data; “brainstorming”; discussing machine learning implementation and interpretability.