RAG

Retrieval-Augmented Generation (RAG) represents an advanced AI technique that combines the strengths of retrieval-based and generation-based models. The methodology involves retrieving pertinent documents from a knowledge base before generating contextually appropriate responses, thereby improving the precision and depth of AI-generated answers.

Simple RAG Flows

Straightforward RAG implementations integrate retrieval and generation processes in the kis.ai ecosystem:

1. Document Storage: The Content block serves as a repository for documents, manuals, and data resources.
2. Retrieval Step: An AI Flow retrieves documents matching user queries, with the AI Gateway directing path selection and model choices.
3. Generation Step: Retrieved information feeds into an AI model through Prompts and the AI Gateway to formulate responses.
4. Output: The final response is a combination of the retrieved information and the generative model’s processing, providing a contextually rich answer.

Complex RAG Flows

Sophisticated RAG implementations incorporate multiple processing layers:

1. Advanced Document Storage and Indexing: Large-scale data management through Content blocks and custom indexing.
2. Multi-Stage Retrieval: Successive refinement of initial retrievals to identify the most applicable documents.
3. Integration with AI Flows: Orchestration of retrieval and generation through AI Flows, employing conditional logic for query-specific handling.
4. Enhanced Generation: Utilize advanced models through the AI Gateway, incorporating A/B testing, load balancing, and optimal response selection from multiple models.
5. Security and Rate Limiting: Protection via access rules and rate limiting at user, team, or route levels.