AI Profiles (Organizations)

Introduction

Welcome to our guide on AI Profiles (Organizations). This feature is designed to help users uncover valuable insights about their competitors, partners, or organizations working within a field of interest. 

Why We Built This Feature

Background

Users come to Cypris to get a clear picture of the state of innovation in a particular field. Commonly, they begin their analysis by identifying which organizations are actively investing R&D time and resources in the space. They want to know (1) Who’s in this market?, (2) What are they working on?, and (3) Where are they going? 

Today, that analysis is being pieced together manually from a wide array of raw data which takes a tremendous amount of time and can lead to missed critical insights. 

Using generative AI and the Cypris database, this feature allows users to automatically generate profiles on organizations found within search results to answer those 3 critical questions. 

‍

User Needs Addressed

• Automatically generate profiles on organizations to uncover:
  • Who they are.
  • What they are working on.
  • Predictions on where they are going.
• Profiles can be downloaded to pdf and shared with colleagues. 

How AI Profiles (Organizations) Works

1. Run any type of search (semantic, boolean, or organization). 
2. Navigate to the organization's tab and identify 1 or many that you’d like to analyze. 
3. Select ‘Generate AI Profile’ and wait 10-15 seconds (we’re working on making this faster).
4. You can view the profile in the side panel or click ‘view profile’ to go to the main company page. 
5. Select the download icon to download the profile to a pdf.

‍

How We Built This Feature

This feature is powered by ‘retrieval augmented generation’ or ‘RAG’. 

Explained simply, RAG is an AI technique that focuses the answers you get from language models by combining external contextual knowledge with text generation. When you ask a question, a RAG model searches through a focused set of documents (or ‘context’) to find the most relevant information. It then uses this information to generate a response, effectively blending up-to-date data with its language skills. This means you get more accurate and context-aware answers, making the AI's responses more accurate and reliable.

‍

Cypris RAG Context for This Feature:

For this feature, we’re primarily looking at the patent portfolio of the company being analyzed alongside some external data retrieved on the entity from our data partner Crunchbase. 

When analyzing just the meta-data (title & abstracts) we currently can look at 600-700 patents. 

When analyzing the full text we can include around 50-60 patents. 

For different prompts, we look at different amounts of context. For example, when a keyword search is done on a company page, we only look at patents that include that keyword to generate the second prompt ‘Describe {Organization’s} work in {user query]’. 

‍