Multimodal Patent & IP Novelty Detector
Develop a cutting-edge AI system for patent and intellectual property (IP) research.. This challenge focuses on building an agentic system that can analyze patent documents (text and images) to determine novelty, identify prior art, and summarize key innovations. The system will leverage GPT-5's advanced reasoning and multimodal capabilities combined with a graph-based RAG architecture. Your solution will involve a custom agent architecture designed to perform specialized tasks: a 'Patent Search Agent', an 'Image Analysis Agent', and a 'Novelty Assessment Agent'. It will employ a graph-based RAG system to interlink patent texts, claims, and associated images, allowing for deeper contextual understanding. MCP-enabled tool integration will facilitate connecting to external patent databases and image recognition APIs, as well as a simulated Getty Images API for multimodal content retrieval and analysis.
What you are building
The core problem, expected build, and operating context for this challenge.
Build a GPT-5 powered multimodal agent system using Graph RAG and MCP for patent and IP novelty detection, integrating text and image analysis.
How work is evaluated
The evaluation will measure the system's accuracy in identifying relevant prior art, its ability to summarize patent innovations effectively, and its performance in detecting novelty using multimodal information.
Challenge datasets
Public datasets, plus the private datasets you have access to.
What you should walk away with
Integrate multimodal capabilities for analyzing patent diagrams and images alongside text.
Implement a graph-based RAG system to connect textual and visual information from patent documents.
Develop MCP-enabled tool integration for accessing external patent databases and image recognition services.
Design a custom agent architecture for patent research, utilizing GPT-5 for core reasoning and summarization.
Build a 'Novelty Assessment Agent' capable of identifying potential prior art and quantifying innovation.
Reference links and supporting material
Sample data for 2 tasks
Challenge setup
The evaluation will measure the system's accuracy in identifying relevant prior art, its ability to summarize patent innovations effectively, and its performance in detecting novelty using multimodal information.
Challenge input
{'new_patent_claims': str, 'key_diagram_base64': str, 'prior_art_dataset': list[dict]}
Agent execution
No agent tool is listed for this challenge.
Evaluated output
{'top_prior_art_ids': list, 'similarity_explanations': dict, 'novelty_score': float, 'multimodal_contribution': float}
- At least one correct prior art patent must be identified for the 'Prior Art Detection' task.
- Innovation summary should be between 50 and 200 words for conciseness.
- Novelty Detection Accuracy target: 0.85
- Python execution harness
Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.
View technical recipe
Configured tools
No tool records are attached.
Evaluation contract
- At least one correct prior art patent must be identified for the 'Prior Art Detection' task.
- Innovation summary should be between 50 and 200 words for conciseness.
Metrics
- Novelty Detection Accuracy · Target 0.85
- Prior Art Recall (RAG) · Target 0.75
- Multimodal Feature Utilization · Target 0.6
- Summary Conciseness & Clarity · Target 0.9
Workflow tasks
- Task 1
Input: {'new_patent_claims': str, 'key_diagram_base64': str, 'prior_art_dataset': list[dict]}
Output: {'top_prior_art_ids': list, 'similarity_explanations': dict, 'novelty_score': float, 'multimodal_contribution': float}
Execution harness: Python
- Task 2
Input: {'patent_document_text': str, 'patent_figures_base64': list[str]}
Output: {'innovation_summary': str, 'key_claims_extracted': list, 'comprehensiveness_score': float}
Execution harness: Python
Recipe freeze marker
No recipe freeze time is recorded.