PromptQL

PromptQL builds a dynamic query plan that can be modified during execution based on intermediate results or explicit instructions from the user. It incorporates control structures (e.g., if-else statements) to handle different scenarios.

This overcomes a rigid retrieval based approach on precomputed embeddings, similarity search, predefined tools, etc, lacking the ability to adjust retrieval strategies dynamically.

DDN provides connectors to seamlessly integrate with any data sources, allowing AI models to access and interact with all your data under a unified schema.

Problem: Traditional RAG systems struggle to integrate data from diverse sources such as SQL databases, NoSQL stores, APIs, and SaaS platforms.

PromptQL has a programmatic runtime environment that allows AI to execute complex computations, data transformations, and statistical analyses. PromptQL handles large datasets by processing data in batches outside the LLM's context window. It can call another LLM within the runtime to analyze text data.

PromptQL enables AI to perform actions and mutations by executing commands that can update data, call APIs, or invoke functions, allowing for end-to-end task completion.

Problem: Many AI applications are limited to data retrieval and cannot perform actions or mutations, such as updating records or triggering workflows.

Problem: LLMs have limited context windows, making it difficult to handle large datasets or maintain conversation context over long interactions, leading to potential loss of information and hallucinations.

Solution: PromptQL uses memory artifacts to store and retrieve information outside the LLM's context window. This ensures consistency, accuracy, and the ability to handle large-scale data without overloading the LLM.

Problem: RAG systems operate like black boxes. When an AI model retrieves information to generate a response, it can’t provide insights into how or why specific data was selected. It is difficult to verify the sources or reasoning behind the LLM's answers.

Solution: PromptQL provides detailed execution plans and can be explicitly asked to explain each step.

Problem: RAG lacks built-in mechanisms for enforcing data access permissions and security protocols. There is always the potential risk of exposure of sensitive data. It cannot tailor data access based on user roles or credentials.

Solution: Hasura DDN integrates fine-grained, role-based access control directly into the data querying process, ensuring that all data access adheres to predefined authorization policies at row, column, and object levels.