LLM Search Query Answers Grounded in Relevant Content Chunks

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional search systems and LLM-based chatbots often provide broad search results that require users to sift through irrelevant information, and LLM-based responses lack context or verifiable sources.

Innovation Solution

A search platform combining semantic search and large language models (LLMs) to identify relevant content, using content embeddings and context-aware prompts to generate answers directly related to user queries.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of information

If conventional search systems are used to provide search results, then users can access information in the content repository, but users must sift through broad and irrelevant information to find answers

Engineering Contradiction:

Improveinformation retrieval completenessVSAvoiduser time to sift through results

Core Design Contradiction:

Loss of informationVSLoss of time

Solution Approach 1:

The system extracts only the most relevant content chunks from the content repository that directly answer the user's query, rather than returning broad search results. The LLM processes these extracted chunks to generate concise answers, eliminating the need for users to sift through irrelevant information while maintaining complete information retrieval.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The LLM acts as an intermediary between the search system and the user. It receives query and relevant content chunks, processes them through semantic understanding, and generates human-like answers that directly address user questions, bridging the gap between raw search results and user comprehension.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Ease of operation

If LLM-based chatbots are used to generate responses, then users receive direct answers to queries, but the responses lack context or verifiable sources

Engineering Contradiction:

Improveuser effort to obtain answersVSAvoidanswer context and verifiability

Core Design Contradiction:

Ease of operationVSReliability

Solution Approach 1:

The system performs preliminary semantic search to identify and retrieve relevant content chunks from the content repository before generating the LLM response. These pre-selected chunks serve as the knowledge base and context for the LLM, ensuring that generated answers are grounded in verified sources while maintaining ease of use.

Inventive Principle:

Principle #10Preliminary action

3Measurement precision

If semantic search is used to identify relevant content, then search accuracy is improved, but the system requires integration with LLMs to generate direct answers

Engineering Contradiction:

Improvesearch result relevanceVSAvoidsystem architecture integration

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The system merges semantic search functionality with LLM-based answer generation into a unified architecture. The semantic search component identifies relevant content chunks, which are then passed to the LLM for answer generation, creating an integrated system that achieves both high search accuracy and direct answer provision without requiring separate independent systems.

Inventive Principle:

Principle #5Merging (Combining)

Data Source

PatentUS12572575B2Using large language models to generate search query answers

Publication Date: 2026.03.10 HIGHSPOT

AI summary

A system generates search query responses based on a content repository that include natural language answers and actions performed based on content items in the repository. The system receives a query associated with the content repository. Based on the query, the system retrieves a set of text chunks that are relevant to the query. At least a portion of the relevant text chunks are sent to a large language model (LLM) to cause the LLM to generate an answer description for the user query, based on the text chunks.