Skip to content

Retrieval Augmented Generation (RAG)

Last reviewed: 7 months ago

Introduction

Retrieval-Augmented Generation (RAG) is an innovative approach in natural language processing that integrates retrieval mechanisms with generative models to enhance text generation.

By incorporating external knowledge from pre-existing sources, RAG addresses the challenge of generating contextually relevant and informative text. This integration enables RAG to overcome the limitations of traditional generative models by ensuring that the generated text is grounded in factual information and context. RAG aims to solve the problem of information overload by efficiently retrieving and incorporating only the most relevant information into the generated text, leading to improved coherence and accuracy. Overall, RAG represents a significant advancement in NLP, offering a more robust and contextually aware approach to text generation.

Examples for application of these technique includes for instance customer service chat bots that use a knowledge base to answer support requests.

In the context of Retrieval-Augmented Generation (RAG), knowledge seeding involves incorporating external information from pre-existing sources into the generative process, while querying refers to the mechanism of retrieving relevant knowledge from these sources to inform the generation of coherent and contextually accurate text. Both are shown below.

Knowledge Seeding

Figure 1: Knowledge seeding
Figure 1: Knowledge seeding
  1. Client upload: Send POST request with documents to API endpoint.
  2. Input processing: Process incoming request using Workers and send messages to Queues to add processing backlog.
  3. Batch processing: Use Queues to trigger a consumer that process input documents in batches to prevent downstream overload.
  4. Embedding generation: Generate embedding vectors by calling Workers AI text embedding models for the documents.
  5. Vector storage: Insert the embedding vectors to Vectorize.
  6. Document storage: Insert documents to D1 for persistent storage.
  7. Ack/Retry mechanism: Signal success/error by using the Queues Runtime API in the consumer for each document. Queues will schedule retries, if needed.

Knowledge Queries

Figure 2: Knowledge queries
Figure 2: Knowledge queries
  1. Client query: Send GET request with query to API endpoint.
  2. Embedding generation: Generate embedding vectors by calling Workers AI text embedding models for the incoming query.
  3. Vector search: Query Vectorize using the vector representation of the query to retrieve related vectors.
  4. Document lookup: Retrieve related documents from D1 based on search results from Vectorize.
  5. Text generation: Pass both the original query and the retrieved documents as context to Workers AI text generation models to generate a response.