Search

US-12619647-B2 - Systems and methods for searching and summarizing financial related documents, and user interface for displaying the same

US12619647B2US 12619647 B2US12619647 B2US 12619647B2US-12619647-B2

Abstract

Systems and methods for searching, summarizing, and automating workflows for financial related documents, and a user interface for displaying the same, are disclosed. The summaries are generated in response to a user query or through natural-language, conversational chat, and they provide the user with underlying supporting documentation to allow the user to follow-up and/or confirm as necessary. An AI-powered document search is performed to identify relevant documents and relevant snippets within the documents. The relevant snippets and/or documents are then summarized by one or more AI models. The models use metadata associated with the documents from the document search to rank the snippets for summarization. AI agents further automate the workflow.

Inventors

  • Rajmohan Neervannan
  • Jaakko Kokko
  • Chris Ackerson
  • Stephen Lynch

Assignees

  • AlphaSense, Inc.

Dates

Publication Date
20260505
Application Date
20240816

Claims (18)

  1. 1 . A system for using one or more artificial-intelligence models to summarize insights across one or more financial documents, the system comprising: a memory storing instructions to be executed by one or more hardware processors; and one or more hardware processors configured to execute the instructions stored in the memory, wherein the instructions, when executed by the one or more hardware processors, cause the system to: receive a first natural-language prompt, wherein the first natural-language prompt is based on input from a user at a remote computing device; use a workflow agent to generate a plan for responding to the first natural-language prompt, wherein the plan comprises a plurality of sub-tasks for execution to generate a response to the first natural-language prompt; use a first large-language model (“LLM”) on the received first natural-language prompt to generate a plurality of sub-prompts to use to execute the plurality of sub-tasks to generate the response to the first natural-language prompt, wherein each of the sub-prompts corresponds to one or more of the sub-tasks; execute one or more searches of one or more datastores based on the one or more sub-prompts, wherein each of the one or more searches identifies a respective plurality of relevant documents based on identified snippets of text within each respective relevant document, and wherein the workflow agent determines, based on the first natural-language prompt, whether to use a single search, a parallel plan, or a sequential plan to execute the one or more searches; use one or more LLMs to generate a response to each of the one or more sub-prompts, wherein each response to the one or more sub-prompts is based on at least some of the respective plurality of relevant documents identified by the search based on the respective sub-prompt; and in response to completion of the plurality of sub-tasks, generate the response to the first natural-language prompt based on the generated responses to the sub-prompts, wherein the response includes at least one LLM-generated summary and one or more citations to the relevant documents used to generate the response.
  2. 2 . The system of claim 1 , wherein the sub-prompts are executed in parallel when it is determined to use the parallel plan, and wherein the responses to the sub-prompts are combined to generate the response to the first natural-language prompt.
  3. 3 . The system of claim 1 , wherein the generated responses to the sub-prompts are combined using a second LLM.
  4. 4 . The system of claim 1 , wherein the sub-prompts are executed sequentially when it is determined to use the sequential plan, such that the generated response to a first sub-prompt is used as an input to generate the response to a second sub-prompt.
  5. 5 . The system of claim 1 , wherein the at least one LLM-generated summary is displayed on a multi-pane user interface.
  6. 6 . The system of claim 1 , wherein the one or more searches include a document search.
  7. 7 . The system of claim 1 , wherein the one or more searches include a vector search.
  8. 8 . The system of claim 1 , wherein the one or more searches include a hybrid search.
  9. 9 . The system of claim 1 , wherein the at least one LLM-generated summary is generated based on one or more topics identified based on the first natural-language prompt.
  10. 10 . The system of claim 1 , wherein the at least one LLM-generated summary is generated based on one or more sectors.
  11. 11 . The system of claim 1 , wherein the one or more data stores are selectable to determine whether a respective data store is included in each of the one or more searches.
  12. 12 . The system of claim 1 , wherein the instructions, when executed by the one or more hardware processors, further cause the system to, based on a selection of one of the citations received from the user identifying a selected relevant document, render on a portion of a user interface one or more of the identified snippets of text associated with the selected relevant document.
  13. 13 . A method for using one or more artificial-intelligence models to summarize insights across one or more financial documents, the method comprising: receiving a first natural-language prompt, wherein the first natural-language prompt is based on input from a user at a remote computing device; using a workflow agent to generate a plan for responding to the first natural-language prompt, wherein the plan comprises a plurality of sub-tasks for execution to generate a response to the first natural-language prompt; using a first large-language model (“LLM”) on the received first natural-language prompt to generate a plurality of sub-prompts to use to execute the plurality of sub-tasks to generate the response to the first natural-language prompt, wherein each of the sub-prompts corresponds to one or more of the sub-tasks; executing one or more searches of one or more datastores based on the one or more sub-prompts, wherein each of the one or more searches identifies a respective plurality of relevant documents based on identified snippets of text within each respective relevant document, and wherein the workflow agent determines, based on the first natural-language prompt, whether to use a single search, a parallel plan, or a sequential plan to execute the one or more searches; using one or more LLMs to generate a response to each of the one or more sub-prompts, wherein each response to the one or more sub-prompts is based on at least some of the respective plurality of relevant documents identified by the search based on the respective sub-prompt; and in response to completion of the plurality of sub-tasks, generating the response to the first natural-language prompt based on the generated responses to the sub-prompts, wherein the response includes at least one LLM-generated summary and one or more citations to the relevant documents used to generate the response.
  14. 14 . The method of claim 13 , wherein the sub-prompts are executed in parallel when it is determined to use the parallel plan, and wherein the responses to the sub-prompts are combined to generate the response to the first natural-language prompt.
  15. 15 . The method of claim 13 , wherein the generated responses to the sub-prompts are combined using a second LLM.
  16. 16 . The method of claim 13 , wherein the sub-prompts are executed sequentially when it is determined to use the sequential plan, such that the generated response to a first sub-prompt is used as an input to generate the response to a second sub-prompt.
  17. 17 . The method of claim 13 , wherein the at least one LLM-generated summary is displayed on a multi-pane user interface.
  18. 18 . The method of claim 13 , further comprising, based on a selection of one of the citations received from the user identifying a selected relevant document, rendering on a portion of a user interface one or more of the identified snippets of text associated with the selected relevant document.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS This application claims the benefit of U.S. Provisional Patent Application No. 63/673,868 filed on Jul. 22, 2024, entitled “SYSTEMS AND METHODS FOR SEARCHING AND SUMMARIZING FINANCIAL RELATED DOCUMENTS, AND USER INTERFACE FOR DISPLAYING THE SAME,” and U.S. Provisional Patent Application No. 63/520,505 filed on Aug. 18, 2023, entitled “SYSTEMS AND METHODS FOR SEARCHING AND SUMMARIZING FINANCIAL RELATED DOCUMENTS, AND USER INTERFACE FOR DISPLAYING THE SAME.” Each of the applications referenced herein are incorporated by reference in their entirety. TECHNICAL FIELD The disclosure relates generally to an architecture for an artificial-intelligence (AI) based market intelligence system that provides an interactive user interface for use with a search engine for searching and summarizing financial related documents. BACKGROUND Conventional web search engines return links to entire documents in response to a search query consisting of keywords or phrases given by the user. In the financial domain, the end user is often a financial analyst who is researching the information source and looking to glean actionable intelligence from financial documents relating a specific company, and a specific industry, or one or more companies. Traditional search methods have provided for natural-language processing and sentiment analysis of financial documents. However, in many cases, these types of analyses have been limited to the specific words and sentences found in the financial documents. More recently, various artificial-intelligence models (e.g., large-language models (LLMs) or machine-learning models) have been used for summarizing documents; however, one problem with using AI models, such as LLMs, to generate summaries is that it is difficult to aggregate existing content to know what information is relevant to the summary being generated. The correct information needs to be fed into the models in order to get valuable and relevant summaries from the models. It is further difficult to generate timely, actionable insights from a volume of new documents. Thus, it is desirable to provide systems and methods for accurately and efficiently summarizing financial documents and generating timely, actionable insights from new documents using artificial-intelligence models, such as LLMs. SUMMARY Systems and methods for searching and summarizing financial related documents, and a user interface for displaying the same, are disclosed. Specifically, a comprehensive artificial-intelligence based market intelligence platform is disclosed. The market intelligence platform described herein combines and integrates the following components: (1) document search powered by one or more artificial-intelligence models; (2) summaries generated using one or more artificial-intelligence models; (3) a freeform, interactive conversational or chat-based user interface powered by one or more artificial-intelligence models for interacting with the document search and the summaries; (4) a private single-tenant enterprise cloud for generating summaries based on proprietary information without leaving the enterprise network; and (5) artificial-intelligence agents for automatically simplifying and performing workflows to surface actionable insights from new information in near real-time or on a scheduled basis. Each of these components is described in more detail below. The market intelligence platform described herein solves two key problems. First, it identifies and summarizes trends across companies, sectors, and markets in real-time or near real-time. Second, it allows users to interact with the documents and summaries in an intuitive, freeform manner. BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 illustrates an example of an implementation of a search system for efficiently conducting contextual and sentiment-aware deep search within a piece of content. FIG. 2 illustrates an overview of the deep search process. FIG. F 3 illustrates examples of the user interface of the deep search system. FIG. 4 illustrates more details of the deep search process in the financial domain. FIGS. 5A and 5B illustrate an example of a user interface for the deep search system for an initial query and a list of results, respectively, in the financial industry. FIG. 6 illustrates an example of a sentiment heat map user interface of the deep search system. FIG. 7 illustrates another example of a search results user interface of the deep search system. FIG. 8 illustrates an example of a search results user interface of the deep search system, where the viewing interface allows the user to compare documents side by side. FIGS. 9A and 9B illustrate portions of a document highlighted that is made possible by the deep search system. FIGS. 10A and 10B illustrate an example of a multi-document summary that is made possible by the deep search system. FIG. 11 depicts a group summary event process flow for both the regularly updated company summaries,