About the AI Context Engine
The AI Context Engine™ is a powerful tool that helps generative AI to better understand the context of your business and data. It utilizes your own business context and data to provide more accurate and insightful responses, leading to better decision-making.
The AI Context Engine achieves this by leveraging data.world's underlying knowledge graph-powered data catalog. Specifically, it is an API-driven application that connects your data and business context (stored in your catalog and the semantic layer through the data.world platform) with Generative AI to enable trusted conversations with structured data. The primary issues addressed by the AI Context Engine revolve around the current limitations of Generative AI and Large Language Models (LLMs), particularly regarding accuracy, explainability, and governance. These models often provide confidently incorrect information, pulling outdated data from presentations or documents without the ability to verify the accuracy or source of the information.
By deploying the data.world AI Context Engine and the data.world Data Catalog platform, you can significantly enhance the accuracy of LLM responses. This enhancement is due to the knowledge graph and semantics, validated by research and the broader data community, making responses over four times more accurate than using SQL alone. The knowledge graph allows you to trace queries back to their origins, showing which triples from the graph were used to generate the response, thereby fostering transparency and trust. Additionally, you have full control over what information is exposed by managing the contents of the knowledge graph.
Key capabilities of the AI Context Engine:
Natural Language Interface: It allows your team to ask complex data questions using natural language and get trustworthy answers at scale.
Tools Deployment: The engine provides APIs, SDKs, and plugins to incorporate tools into your preferred AI applications and chat interfaces.
Audit Trace and Tools: It comes with full audit and traceability provision along with natural language explanations for sensitive data recovery.
Automated Agents: The engine uses agents for key tasks such as glossary lookups, query validation, and query explanation.
The AI Context Engine connects the dots between your data and your business context like terms, definitions, metrics, and processes. This allows AI-powered applications to provide real analytics and business decision support. It also ensures governance, stopping hallucinations of generative AI by providing explainable, governable responses. It helps everyone understand how AI reached an answer, making AI responses auditable and reusable for future queries.
Terms of use
The AI Context Engine is Built with Meta Llama 3. By using the AI Context Engine you and your users agree to abide by https://llama.meta.com/llama3/license/ . You and your users also agree that you are solely responsible and liable for the use or distribution of your inputs and the outputs generated by the AI Context Engine, including for any claim by any third party arising out of the use or distribution of such inputs and outputs against data.world. Furthermore, you agree that you will not engage in or facilitate any action or generate any content that infringes, misappropriates, or otherwise violates any third-party rights, including the outputs or results of any products or services using the AI Context Engine.
High-Level Overview of the AI Context Engine Architecture
The AI Context Engine is designed with an API-first philosophy, seamlessly integrating into various applications and workflows. This versatile tool can function as a custom LLM tool, a LangChain (or equivalent) tool, or be called directly providing a standardized interface for capitalizing on the engine's capabilities. The role of a Data Catalog is of pivotal importance acting as a central repository of metadata about utilized data sources.
Much like the data.world platform, AI Context Engine derives its power from its carefully structured input and processes. In data.world, metadata forms the core, offering details about data like location, structure, quality, usage, relationships, meaning, and lineage. AI Context Engine™, too, benefits from a similar, structured collection of data, particularly in ensuring the delivery of accurate and governed responses.
Moreover, just as data connections in data.world interact with the data through metadata collection, virtual queries, and data extraction, the AI Context Engine's capabilities are critically dependant on the various means of interaction with the data.
The AI Context Engine™'s architecture is structured into several primary layers, all of which can be accessed through the API:
API Gateway: This is the access point for all interactions with the AI Context Engine. It handles tasks like authentication, authorization, rate limiting, and others, similar to the API functions detailed in the Developer docs.
Query Processing: This layer processes user queries acquired through the API. It mirrors the advanced query handling described in data.world’s API query guide, utilizing natural language processing techniques to comprehend the query's intent and identify relevant information.
Semantic Layer: This layer utilizes semantic web technologies for structured representation of knowledge and data, similar to how data.world uses RDF (Resource Description Framework) for data presentation. The custom OWL ontology used here is based on your Data Catalog and defines the relevant concepts, relationships, and constraints for your domain.
Data Virtualization Layer: This layer integrates data from a variety of sources, such as databases, APIs, and cloud storage, reminiscent of the various sources supported by data.world. It uses R2RML mappings to convert structured data into RDF, ensuring seamless integration with the semantic layer. Your Data Catalog plays a vital role here by enabling accurate data source interpretation and querying by the AI Context Engine.
LLM Interaction: This layer interacts with the LLM (Mixtral) to generate responses. This interaction is powered by the structured data and knowledge representation from the Semantic and Data Virtualization layers. The Data Catalog may also supply metadata to enrich the context and guide LLM's response generation.
Response Generation: The final response to user queries is created in this layer, with potential post-processing of LLM's output to optimize accuracy, relevance, and policy compliance akin to the custom SQL queries described by data.world. The Data Catalog can further enhance the response quality by supplying additional information or validation.
Integration Capabilities
The AI Context Engine's API-first design makes it highly adaptable to various integration scenarios:
Custom LLM Tool: You can easily integrate the AI Context Engine™ into your own custom LLM tool, allowing you to leverage its advanced capabilities for semantic understanding, data integration, and response generation.
LangChain (or Equivalent) Tool: The API can be seamlessly integrated with LangChain or similar frameworks, enabling you to incorporate the AI Context Engine™ into complex LLM-powered workflows.
Direct API Calls: For maximum flexibility, you can directly call the AI Context Engine™ API from your applications, giving you full control over the integration process.
The AI Context Engine Starter Kits allow you to jumpstart your custom application development using the AI Context Engine. These kits are delivered through GitHub repositories and provide all the necessary code and configuration items to get you up and running with a new Generative AI experience in no time.
The Starter Kits are available for various use cases, including creating chatbots for MS Teams, integrating AI insights into Slack, and developing data-rich web applications with Streamlit. This provides users direct access to data insights within these frequently used tools and it simplifies Q&A interactions with structured data.