AI EXPERTISE

Patent & IP Subject Matter Expertise for AI Training and Evaluation

Patent law is unusually demanding for AI systems. A response can sound fluent and still be wrong because it overlooks a claim limitation, misreads a technical relationship, relies on an unsupported legal proposition, applies the wrong jurisdictional standard or confuses what a reference actually discloses with what might be obvious to add.

CrossVision IP provides senior practitioner-level patent and intellectual-property expertise for AI training, evaluation and quality assurance, with particular strength in Chinese patent law, CNIPA examination, patent claims, prior art, cross-border prosecution and Chinese-English patent materials.

Led by Leo Liu, a Chinese Patent Agent with 24+ years of hands-on patent practice, the work focuses not merely on whether an AI answer sounds reasonable, but on whether it is legally, technically and evidentially supportable.

Discuss an AI Expert ProjectAbout Leo LiuDownload Resume (PDF)
AI TRAINING & EVALUATION

What I Can Support

Patent-domain AI evaluation requires both legal judgment and technical reading. CrossVision IP supports model-training and evaluation workflows where the quality of patent reasoning must be assessed by an experienced practitioner.

01

AI Response Evaluation

Evaluate AI-generated patent and intellectual-property outputs for legal accuracy, factual accuracy, technical consistency, completeness of reasoning, treatment of claim limitations, use of prior art, evidentiary support, jurisdiction-specific accuracy, unsupported assumptions and hallucinated conclusions.

  • Compare alternative model responses
  • Score legal and technical correctness
  • Identify missing claim limitations
  • Flag unsupported assumptions
  • Detect hallucinated facts or authorities
  • Explain why an answer is correct, incomplete or unsound
02

Expert Question & Task Design

Develop practitioner-level questions and evaluation tasks that test whether an AI system can reason through real patent problems rather than reproduce generic legal explanations.

  • Patentability, novelty and inventive step / obviousness
  • Claim interpretation and prior-art mapping
  • Claim drafting and Office Action analysis
  • Infringement, validity and FTO
03

Gold Answers & Reference Responses

Prepare or review gold-standard reference answers that identify the governing framework, the relevant claim and technical facts, the analytical steps connecting those facts to the rule and the conclusion actually supported by the record.

  • Reference-answer preparation and expert validation
  • Chinese- and English-language patent questions
  • Multi-step reasoning
  • Evidence-based conclusions
04

Evaluation Rubrics

Develop structured evaluation criteria that distinguish between a correct answer, a correct conclusion reached through defective reasoning, an incomplete answer, a jurisdictionally incorrect answer and a hallucinated or unsupported response.

  • Legal and technical accuracy
  • Claim analysis and evidence use
  • Reasoning quality
  • Jurisdictional fit
PATENT-SPECIFIC EVALUATION

Where Patent AI Commonly Fails

Patent analysis creates failure modes that are difficult to detect without substantial prosecution and claim-analysis experience. The following areas are particularly suitable for expert evaluation.

Novelty

Evaluate whether an AI system correctly determines whether every limitation of a claim is disclosed in a single prior-art reference.

  • Combining references during novelty analysis
  • Treating similarity as disclosure
  • Relying on unsupported implicit disclosure
  • Overlooking dependency or claim relationships
  • Misreading patent drawings
  • Importing features not actually disclosed

Inventive Step / Obviousness

Assess whether the model correctly analyzes distinguishing features, technical effects, technical problems, motivation to modify or combine references, reasonable expectation of success and hindsight reasoning.

For Chinese patent matters, evaluation can address the CNIPA framework for identifying the closest prior art, distinguishing technical features, technical effects and the technical problem actually solved.

For U.S. and European matters, evaluation can distinguish the different analytical approaches used in USPTO and EPO practice.

Claim Drafting & Claim Quality

Review AI-generated claims for technical completeness, internal consistency, claim hierarchy, antecedent basis, dependency, unsupported generalization, omitted relationships, inconsistent terminology and consistency with the specification and drawings.

The issue is not merely whether the claim is grammatically correct, but whether it accurately captures the disclosed invention and functions as a meaningful patent claim.

Prior-Art Analysis

Evaluate whether the model correctly distinguishes what a reference actually discloses from what can only be inferred or added through later reasoning.

  • Explicit versus inferred disclosure
  • Feature-to-paragraph mapping
  • Drawing interpretation
  • Similar terminology with different structures
  • Reference combinations
  • Unsupported technical mappings

Office Action & Prosecution Analysis

Evaluate AI-generated analysis of CNIPA examination opinions, USPTO Office Actions, EPO examination communications, cited references, amendments, applicant arguments and prosecution history.

  • Identify the actual rejection or objection
  • Distinguish legal from drafting issues
  • Evaluate cited-reference reasoning
  • Verify proposed amendment support
  • Identify new-matter / support risks
  • Assess whether arguments address the examiner's actual position
INFRINGEMENT · VALIDITY · FTO

AI Evaluation Beyond Patent Prosecution

Patent AI systems may also be asked to analyze issued claims against products, prior art or technical evidence. These tasks require disciplined separation between what the claim requires, what the product or reference actually contains and what the available evidence establishes.

Infringement

  • Claim-element analysis
  • All-elements comparison
  • Accused-product feature mapping
  • Literal infringement
  • Technical equivalence considerations
  • Evidence-based claim charts

Validity

  • Prior-art mapping
  • Novelty
  • Inventive step / obviousness
  • Support and disclosure issues
  • Prosecution-history review
  • Invalidity-risk reasoning

Freedom to Operate

  • Product-feature analysis
  • Claim screening
  • Patent-family review
  • Legal-status review
  • Risk prioritization
  • Design-around analysis
CHINESE PATENT LAW & CNIPA

A Distinctive Jurisdictional Capability

Many legal AI systems have extensive English-language legal data but substantially less practitioner-level coverage of Chinese patent law and CNIPA examination practice.

Leo has worked in Chinese patent practice since 2002 and obtained his Chinese Patent Agent qualification in 2004. His experience enables evaluation not only of general patent-law reasoning but of whether a model accurately reflects Chinese examination practice and Chinese-language patent materials.

Chinese Patent LawCNIPA Patent Examination GuidelinesInvention PatentsUtility ModelsDesign PatentsNoveltyInventive StepClaim DraftingClaim AmendmentsClaritySupportSufficiency of DisclosureReexaminationInvalidationInfringementPCT National Phase
CHINESE ↔ ENGLISH

Multilingual Patent Evaluation

Patent translation is not ordinary translation. A technically fluent sentence may still alter claim scope, introduce unsupported meaning, omit a relationship or create inconsistency with the original disclosure.

CrossVision IP can support AI evaluation involving Chinese and English patent specifications, bilingual claims, CNIPA documents, USPTO and EPO materials, technical terminology and prosecution correspondence.

Translation Fidelity

Has the meaning of the source patent document been preserved?

Patent-Law Effect

Does the translation alter technical or legal scope?

Terminology Consistency

Is the same technical concept translated consistently throughout the document?

Unsupported Addition or Omission

Has the AI introduced meaning that does not exist in the source or omitted information that affects the analysis?

HUMAN EXPERT VALIDATION

Why Senior Expert Review Still Matters

Patent AI often fails in ways that are difficult to identify without substantial patent experience. A response may use the right legal vocabulary and still reach an unreliable conclusion.

Senior expert review provides a validation layer between fluent AI output and professionally reliable patent analysis.

  • A correct rule applied to the wrong facts
  • A technically plausible interpretation unsupported by the patent
  • A missing claim limitation
  • An incorrect inference from a figure
  • An inventive-step argument based on hindsight
  • An amendment that creates a support problem
  • A translation that subtly changes claim meaning
  • A confident citation that does not support the proposition
  • A technically similar reference mapped to the wrong claim feature
ENGAGEMENT TYPES

Typical AI and LegalTech Engagements

AI Training Projects

Creation and expert review of patent and intellectual-property training materials.

Model Evaluation

Expert scoring, comparison and qualitative review of model responses.

Benchmark Development

Creation of practitioner-level patent questions and benchmark tasks.

Gold-Answer Development

Preparation and validation of reference responses for expert-level patent datasets.

Rubric Development

Definition of scoring criteria for legal, technical and evidentiary reasoning.

Reasoning Red-Teaming

Identification of failure modes, unsupported reasoning and hallucinations in patent-related AI outputs.

LegalTech Product Validation

Domain review of AI-powered patent tools, research workflows and multilingual outputs.

Areas of Subject-Matter Expertise

Patent Law

Chinese Patent Law · CNIPA Examination · Patentability · Novelty · Inventive Step / Obviousness · Claims · Prior Art · Patent Prosecution · Reexamination · Invalidation · Infringement · FTO · Validity

Cross-Border Practice

CNIPA · PCT · USPTO Support · EPO Support · China-to-U.S. Patent Practice · China-to-Europe Patent Practice · Foreign Applications Entering China

Technical Fields

Mechanical Systems · Electromechanical Products · Consumer Products · Medical Devices · Battery & Energy Systems · Optics & Displays · Electronics · Communication Technologies · Software-Related Inventions · Manufacturing Technologies

Languages

Chinese — Native
English — Professional Working Proficiency

PROFESSIONAL FOUNDATION

AI Evaluation Built on Actual Patent Practice

CrossVision IP's AI evaluation capability is built on hands-on patent practice rather than general legal-data annotation.

24+ years in intellectual-property practiceChinese Patent Agent qualification since 2004Senior roles in Chinese patent agenciesCross-border patent prosecution experiencePatent infringement and validity analysisTechnical-investigation workIP dispute mediationProfessional writing and training
About Leo Liu

Who I Work With

AI Companies & Model Teams

For patent and Chinese-law domain evaluation, training and expert review.

AI Data & Expert Platforms

For high-skill legal evaluation, benchmark tasks, reference answers and rubric development.

LegalTech Companies

For patent-domain validation, workflow review and multilingual product evaluation.

Expert Networks & Recruiters

For project-based Patent, IP, CNIPA and Chinese Legal SME engagements.

WORK WITH LEO

Looking for Patent or Chinese-Law Expertise for an AI Project?

CrossVision IP is available for remote and project-based engagements involving patent AI training, legal AI evaluation, Chinese patent law, CNIPA expertise, intellectual-property datasets, multilingual legal review and LegalTech product validation.

Based in Shenzhen, China · Available for Global Remote Collaboration
Discuss an AI Expert ProjectAbout Leo Liu