Skip to main content

IPRally Lite User guide: Free-Text Search and Image search

Guide for IPRally Lite users - including Best Practice

JP - John Paul Keeler avatar
Written by JP - John Paul Keeler
Updated over 3 weeks ago

Free-Text Search User Guide - Lite

This guide covers the basic steps of carrying out a Free-Text search for an invention, when you have an Invention Disclosure/Project Scope Description; Image or any other Technical Description. It will also cover what is not possible to do with Free-Text.

English is the native language of IPRally and optimal as input. German, French, Spanish, Chinese, Japanese, Italian, Dutch, Danish, Finnish and Swedish are supported via machine translation (with a maximum 5000 characters). The translation happens inside IPRally, making it secure.

Free-Text Search takes your input (see above) and curates a Knowledge Graph for searching. The results are shown by ranking (using the AI Score)

Free-Text Search Is Not a Generative Ai Search - meaning, you must describe the object(s) and features you are searching for, and NOT a 'ChatGPT' style prompt expecting the Ai engine to curate a more robust description (with potential hallucination).

The AI Score

The maximum score is 100 and it describes the overall graph similarity. The score is calculated as a vector distance between the query graph and prior art graph after AI processing of the graphs. With short (e.g. claim) query graphs, no scores higher than 60 are usually reached as there are many irrelevant features producing noise in the prior art graphs.

FAQ's/Best-Practice for Free-Text Searching

When using the Free-Text tool, please use the below insights to drive your searches:

  1. The AI cannot (yet šŸ˜Š) read your mind: Therefore, give it enough

    • Context information:

      • E.g. ā€œdeviceā€ -> ā€œmobile communication deviceā€ / ā€œmethodā€ -> ā€œmethod for processing signalsā€

    • Technical details:

      • Detailed is better than general

      • Functional relationships are important (and possible!) in IPRally

      • E.g. ā€œelongated member engages with circular member to oscillate pendulumā€

  2. The AI is trained with real claims and specifications:

    • Donā€™t be afraid of ā€œpatent jargonā€ (e.g. ā€œmeans for ā€¦ā€ / ā€œfirst elementā€): The AI understands it!

    • Claims-level of details is a good starting point

  3. In free-text search, use natural and consistent language:

    • Full sentences, include articles, internally coherent part names, etc.

  4. Common abbreviations are recognised but often longer format is preferred

    • e.g. ā€œSEMā€ -> ā€œscanning electron microscopeā€

Examples*:

Good: "The invention is a vehicle-to-vehicle (V2V) communication system for autonomous vehicles, using both wireless technologies (WiFi, LTE, 5G) and RFID. It features a dual-module framework that dynamically switches between high-bandwidth and low-bandwidth communication modes to optimize data transmission. This system enhances safety, efficiency, and passenger interaction in autonomous vehicles."

Poor: "A V2V system for autonomous vehicles using wireless tech and RFID to enhance safety."

Why? The good example has technical details around the objects, and how they interact to deliver the function being searched. There are functional relationships described, and the language is descriptive vs. prompting. The poor example is too brief, lacking functional relationships, and doesn't describe the 'novelty' well enough. The results will be too 'vague' compared to the good input.

What not to Do:

Don't use a 'ChatGPT' style Generative Ai Prompt with Free-Text search.

Please...

It will greatly diminish the ability for our Graph-Ai to understand the subject, and therefor provide lower-quality results.

Example of a poor 'Prompt' inspired search input:

"A V2V system for autonomous vehicles using wireless tech (WiFi, LTE, 5G) and RFID to enhance safety. Focus on highlighting the dual-module communication approach that integrates high-bandwidth wireless technologies for real-time data and low-bandwidth RFID for stable communication. Emphasize how the system dynamically switches between these modes to optimize performance, targeting improvements in safety, efficiency, and passenger experience in autonomous vehicles."

Why? Our Graph Ai is not a LLM (As all GPT's are), but rather a Knowledge Graph Neural Network. This requires the input to be similar to the patent documents being searched, and does not operate as an LLM would in 'filling in the blanks'

* (Curated using ChatGPT 4.o - resemblance to any prior art is by chance)

Image search

Image Search lets you find patents by simply uploading an image. Unlike traditional image searches that focus on visual similarity, Image Search focuses on the concepts, technologies, and inventions depicted in the uploaded image. It finds patents describing similar ideas, even if they look different visually or have no images at all.

Using Image Search

1. Accessing Image Search

To start an Image Search, navigate to the IPRally homepage and select "Image search" from the list of search options. If AI Assistant Features are not enabled, you'll see a message about it instead of the search functionality.

2. Uploading an Image

Click on the upload area or drag and drop your image file. Supported file types are JPEG, PNG, or WEBP, with a maximum file size of 20MB.

3. Adding Context (Optional)

Below the image upload area, you'll find an optional text field where you can add context for the image. Use this to provide additional information or instructions for the AI. The search works with just the image, but adding context can help guide the AI's interpretation of the image. You can:

  • Describe non-visible elements. Example: "Exhaust gas catalyst structure"

  • Focus on specific features. Example: "Focus on the cooling mechanism"

  • Clarify overlooked details. Example: "Blue parts are ceramic composite"

  • Provide labels for diagrams. Example: "1: Rotor, 2: Stator, 3: Bearing"

4. Searching Patents

After uploading your image and optionally adding context, click the "Search Patents" button to initiate the search.

5. Search Results and Refinement

View the list of patent documents matching your image search. You can edit the context text if the results do not focus on the correct information.

Best Practices for Image Search

  1. Use clear, detailed images when possible.

  2. Remember that Image Search finds conceptually related patents, not visually similar images. The results may not look like your uploaded image.

  3. If the AI misinterprets the image, provide additional context to clarify the aspects you're interested in. Don't hesitate to run multiple searches, adjusting your input each time.

  4. Edit the search graph directly for more precise control over the technological features you want to search for.

  5. If you want to search with a different image, start a new search case. You cannot change the uploaded image after initiating a search.

Technical Process

IPRally uses Claude 3.5 Sonnet, a Large Language Model (LLM), to analyze the uploaded image and convert it into a textual description. This description is then transformed into a search graph, which is a structured representation of the invention's key features and relationships. The search graph is then used to find relevant patent documents, similar to how graph search and text search function in IPRally.

Enabling Image Search

If you are an administrator and want to enable AI Assistant Features (including Image Search) for your organization:

  1. Go to Settings

  2. Navigate to "Licenses" under the "Company" header

  3. Review and accept the Generative AI Terms

  4. Enable AI Assistant Features

Did this answer your question?