{"id":40355,"date":"2025-09-28T17:12:08","date_gmt":"2025-09-28T15:12:08","guid":{"rendered":"https:\/\/www.dbi-services.com\/blog\/?p=40355"},"modified":"2025-09-28T17:12:10","modified_gmt":"2025-09-28T15:12:10","slug":"rag-series-naive-rag","status":"publish","type":"post","link":"https:\/\/www.dbi-services.com\/blog\/rag-series-naive-rag\/","title":{"rendered":"RAG Series – Naive RAG"},"content":{"rendered":"\n

Introduction<\/strong><\/h2>\n\n\n\n

Since my last series on pgvector I had quite a fun time to work on RAG workflows on pgvector and learned some valuable lessons and decided to share some of it in a blog post series on the matter.
We will discover together where all RAG best practices are landing for the past 2 years and how can as a DBA or “AI workflow engineer” improve your designs to be production fit.
We start this series with Na\u00efve RAG, this is quite known but important and foundational for the next posts of this series.<\/p>\n\n\n\n

What is Retrieval-Augmented Generation (RAG)?<\/strong><\/h2>\n\n\n\n

Retrieval-Augmented Generation (RAG)<\/strong> is a technique that combines the power of large language models (LLMs) with information retrieval. Instead of relying solely on an LLM\u2019s internal knowledge (which may be outdated or limited, and prone to hallucinations), a RAG system retrieves relevant external documents and provides them as context for the LLM to generate a response. In practice, this means when a user asks a question, the system will retrieve<\/strong> a set of relevant text snippets (often from a knowledge base or database) and augment<\/strong> the LLM\u2019s input with those snippets, so that the answer can be grounded in real data. This technique is key for integrating businesses or organizations data with LLMs capabilities because it allows you to implement business rules, guidelines, governance, data privacy constraints…etc.
Na\u00efve RAG is the first logical step to understand how the retrieval part works and how it can impact the LLM output. <\/p>\n\n\n\n

A RAG pipeline typically involves the following steps:<\/p>\n\n\n\n

    \n
  1. Document Embedding Storage<\/strong> \u2013 Your knowledge base documents are split into chunks and transformed into vector embeddings, which are stored in a vector index or database.<\/li>\n\n\n\n
  2. Query Embedding & Retrieval<\/strong> \u2013 The user\u2019s query is converted into an embedding and the system performs a similarity search in the vector index to retrieve the top-$k$ most relevant chunks.<\/li>\n\n\n\n
  3. Generation using LLM<\/strong> \u2013 The retrieved chunks (as context) plus the query are given to an LLM which generates the final answer.<\/li>\n<\/ol>\n\n\n\n

    Try It Yourself<\/strong><\/h2>\n\n\n\n

    Clone the repository and explore this implementation:<\/p>\n\n\n

    \ngit clone https:\/\/github.com\/boutaga\/pgvector_RAG_search_lab\ncd pgvector_RAG_search_lab\n<\/pre><\/div>\n\n\n
    The lab includes:<\/p>\n\n\n\n