Enterprise Knowledge Management

Key Business Use Cases: Enterprise Knowledge Management and Search

Knowledge Management Capabilities

Storing and managing knowledge
Content and knowledge management is fully supported and includes the ability to go straight from utterance to knowledge as well as addressing the ongoing workflows to manage knowledge. We recently deployed an approach for Q&A management that is made up of three main areas of functionality: combining LLMs with Elastic Search capability, and a heavy focus on knowledge workflow management. LLMs, external connectors to Q&A search tools, encoder NLU, graph based knowledge management are all a part of the toolkit that customers can use to handle simple to complex management of knowledge. We encourage our customers to use LLM decoders in combination with a graph DB and taxonomy to create a hypergraph with knowledge domains that point to unstructured data documents.

OneReach.ai Lookup Product
OneReach.ai’s Lookup offering allows customers to supplement their external knowledge sources with a vector-based approach inclusive of using LLMs to generate metadata for each embedding.

Step 1 – Import: Data is uploaded from a vast array of sources (acceptable formats include: .csv, .pdf, .html, .txt, and files from Notion, web pages, etc.) to create what we call a Collection of data.

Step 2 – Analyze and Refine: The Collection of data can be augmented manually or automatically via Properties (metadata and tags). This can be automated in agentic workflows or done directly in the Lookup UI.

Step 3 – Answer and Surface: When users build AI agents and solutions that are given permission to access specific Collections of knowledge, they can easily answer and surface questions, browsing content and providing the most relevant responses via Agentic RAG capabilities.

Analyze and compare documents
The Lookup product provides advanced architecture that also allows users to analyze and compare documents in a Collection. For example, users can search, “Compare commercial papers at the end of each year from 2015 to 2022” and the AI agent will return comparisons of the documents with similarities and differences clearly noted.

Optical Character Recognition (OCR)
Because we have an open architecture platform, we integrate with any OCR systems available on the market. This allows custom and out-of-the-box OCR features.

Manage edits and multiple versions
When knowledge redundancy makes standard vectorization or summarization impractical, we utilize graph databases to store information ontologically. We point to summarizations and descriptions that point to the source document, and the source document can be changed as knowledge is updated, without having to re-vectorize the system.

Managing conflicting information
One methodology is via graph DB ontology, breaking up knowledge into domains, and assigning domain owners and permissions. Once we assign certain doc owners and permissions to knowledge, we can run scripts every time knowledge is added and show clusters of vectors, with information from the author like date, location, etc. We are able to surface any redundant information to help manage conflicting content, to make sure that there is only one vectorized passage pointing at a source for information.

Agentic RAG
Agentic RAG is an advanced form of RAG that uses AI agents to improve understanding of the query and improve retrieval accuracy. The AI agent dynamically adapts the retrieval process based on the context and intent of user queries, improving the accuracy of the response. Traditional search architecture works well with structured data, and agentic RAG helps overcome limitations of some forms of structured data and especially unstructured data. Our Lookup Product provides Agentic RAG capabilities to optimize enterprise search capabilities.

Graph RAG
Graph Retrieval-Augmented Generation, is an advanced information retrieval approach that enhances retrieval-augmented generation (RAG) by utilizing knowledge graphs instead of vector databases. This method provides LLMs with structured, context-rich information, improving the accuracy and relevance of generated responses. By organizing data in a graph structure, Graph RAG captures complex relationships and hierarchies, allowing for more sophisticated reasoning and inference, and helping AI agents mitigate bias. By leveraging these graph-based relationships, GraphRAG can trace provenance, balance information representation, and prioritize counterfactuals or underrepresented viewpoints, reducing the risk of biased outputs while maintaining explainability.

Integrate with external knowledge bases
GSX integrates with many commercial knowledge sources which allows customers to inherit their legacy tools while also benefiting from our comprehensive knowledge management and access capabilities. However, the quality of knowledge in these legacy tools varies dramatically and is rarely designed for generative AI use cases (e.g. lack of metadata, not using vector embeddings, etc.).

Search Capabilities

Semantic Search
OneReach.ai provides natively managed services across the entire Semantic Search Stack. Users can use a combination of embeddings, vector databases, graph databases, and LLMs to create semantic understanding of your enterprise data at scale.

• Indexing: Vector databases and graph databases can be used to index the content of documents with semantic similarity, leading to more semantically relevant results than traditional text-based indexing methods can produce.
• Querying: Generate more natural and informative search queries in combination with a personalized biosketch and other available metadata. This can help users to find the information they are looking for more easily
• Retrieval: Rank documents more accurately and relevantly. This can help users to find the most useful information more quickly.
• Evaluation: Evaluate the effectiveness of information retrieval systems more accurately. This can help researchers to improve the design and implementation of these systems.

Traditional Search
We also offer traditional search capabilities, using keywords to create queries and find relevant results. This is a fast, cheap and reliable way to perform search that, depending on your use case, may be all you need to achieve strong information retrieval results.

Hybrid Search
Often, the best information retrieval results follow from a hybrid approach combining traditional keyword search and semantic search. With this hybrid approach, traditional search acts as the first layer to quickly and cheaply narrow the search to a localized domain. Semantic search is then applied over this smaller domain, resulting in an improved search outcome for the user.

Gather feedback from users
Feedback from users, such as simple thumbs up or down, can be used to reinforce model performance or to provide more detailed feedback on the relevance of search results. LLMs can also be used to generate their own success metrics, such as the number of times a human is requested for support or the amount of time a user spends reading a document. This feedback can be used to improve the performance of the information retrieval system.

LLM Custom Prompt Engineering

OneReach.ai’s flexible information retrieval capability extends to custom prompting methods. There are infinite ways to optimize retrieval through prompting, and new techniques are being discovered continuously.

Chain-of-Thought (CoT)
These prompts ask the LLM to explain its reasoning step-by-step, which can help to identify important concepts and relationships in the query.

Hypothetical Document Embeddings (HyDE)
These prompts ask the LLM to imagine a document that would answer the query, and then to generate a vector representation of that document. This vector can then be used to retrieve similar documents from a corpus.