Genomic Patent Prior Art and Infringement Classifier with AutoGen
IP disputes in single-cell sequencing require rigorous multi-agent analysis of patent claims against scientific publications. Build a conversational multi-agent audit team using Microsoft AutoGen to parse claim charts and highlight overlapping method steps. Achieve 90% precision on patent claim overlap extraction.
What you are building
Orchestrate a multi-agent patent audit system using Microsoft AutoGen to evaluate genomic tech claims against prior art.
How work is evaluated
Evaluates accuracy in detecting matching elements between single-cell patent claims and prior art protocols.
Datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
risk_level_format
Risk level must be HIGH, MEDIUM, or LOW.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
claim_overlap_precision
Precision of identified claim limitations matching prior art text • target: 0.9 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Learning goals
Implement multi-agent conversation topologies using Microsoft AutoGen framework
Extract structured patent claim elements from USPTO documentation
Cross-examine genomic library preparation protocol steps against prior art literature
Generate evidence reports highlighting potential claim infringement and anticipation risks
Resources
Dataset of 50 paired single-cell RNA-seq patent claims (e.g. 10x Genomics vs Parse) and corresponding research paper methods.
Challenge setup
Evaluates accuracy in detecting matching elements between single-cell patent claims and prior art protocols.
Challenge input
JSON containing patent_claim_text and prior_art_method_text.
AutoGen
Framework specified for multi-agent group chat consensus.
Evaluated output
JSON object with overlapping_elements array, infringement_risk_level, and consensus_reasoning.
- Risk level must be HIGH, MEDIUM, or LOW.
- Precision of identified claim limitations matching prior art text • target: 0.9 • range: 0-1
- Benchmark: Biotech Patent Claim Infringement Benchmark
- Claim Overlap Precision target: 0.9
- 1 public reference case
- Python execution harness
- Python sandbox (unavailable on Versalist)
Shows the first task, up to three tools, and up to three evidence items.
View technical recipe
Configured tools
- AutoGen · Required
- HeyBoss AI · Optional
- Qwen2.5-72B · Optional
Evaluation contract
- risk_level_format · Weight 1
- claim_overlap_precision · Weight 1
Metrics
- claim_overlap_precision · Target 0.9
Benchmarks
- Biotech Patent Claim Infringement Benchmark
- biotech
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON containing patent_claim_text and prior_art_method_text.
Output: JSON object with overlapping_elements array, infringement_risk_level, and consensus_reasoning.
Execution harness: Python
Evaluation tools
- AutoGen · Multi-agent Conversational Framework
Recipe freeze marker
No recipe freeze time is recorded.