Skip to main content
The Vector Store RAG template enables you to build a retrieval-augmented generation (RAG) system that grounds AI responses in your own documents using a vector database, optimized prompting, and dynamic execution. This advanced setup combines vector-based knowledge retrieval with prompt optimization and the LLMC Executor to deliver consistently high-quality, context-grounded answers.

​
Prerequisites

  • OpenAI API Key (required for both embeddings and model)
  • A document file (PDF, TXT, MD, CSV, JSON, DOCX, etc.)
  • Basic familiarity with LLMC components

​
Create the Vector Store RAG Flow

​
1. Load Your Documents

  • Start with the Load Data Flow (top-right section of the workspace).
This flow prepares your knowledge base for retrieval. Steps:
  • Upload your document using the File component.
  • Split the content into manageable text chunks via Split Text.
  • Generate embeddings with OpenAI Embeddings (text-embedding-3-small).
  • Store embeddings in LLMC Vector DB for future semantic retrieval.
Ensure your OpenAI API Key is added to the Embeddings component.

​
2. Ask Questions Using RAG

Once your data is embedded, switch to the Retriever Flow (bottom section). Components:
  • Chat Input - Enter your question in the Playground.
  • OpenAI Embeddings - Converts your query into a vector for semantic search.
  • LLMC Vector DB - Retrieves relevant document chunks based on your query.
  • Parser - Formats and cleans retrieved content for the LLM.
  • Prompt - Combines the retrieved context with your question into a structured prompt.
  • LLMC Executor - Executes the prompt using the selected model and returns the answer.
  • Chat Output - Displays the final response in the Playground.
Try different search queries to see how the retrieved context changes.

​
3. Optimize Your Prompts

Use the Prompt Optimizer Flow (top-left section) to refine your prompt instructions and improve response quality. How it works:
  • Automatically generates prompt variations and test cases.
  • Scores and ranks are prompted by relevance, accuracy, and tone consistency.
  • Displays a leaderboard of best-performing prompts.
To run it:
  • Input your optimization task (e.g., “Answer document questions accurately and concisely”).
  • Click Run on the Results node.
  • Review and select the top-performing prompt from the leaderboard.

​
4. Execute and Compare Results

Once optimized prompts are ready, use the LLMC Executor to run them. In the LLMC Executor:
  • Select your optimized Prompt and preferred Model (gpt-4o, claude-4-sonnet, etc.).
  • Paste your OpenAI API Key if not already configured.
  • Click Play to execute the query.
  • The output displays your final, context-grounded answer.

​
Modify or Extend

  • Change the prompt template in the Prompt component to control how context and questions are presented.
  • Swap models in LLMC Executor to test performance differences.
  • Adjust chunk size and overlap in Split Text for better retrieval precision.
  • Use Results Leaderboard to continuously improve response quality over time.
  • Create multiple collections in LLMC Vector DB to organize different knowledge bases.

​
Configuration Checklist

ComponentConfiguration
File / Split TextUpload and preprocess document data
OpenAI EmbeddingsEmbedding model: text-embedding-3-small
LLMC Vector DBStores embedded vectors for retrieval
Prompt Optimizer FlowGenerates and ranks prompt variants
LLMC ExecutorRuns the selected model and prompt
API KeyRequired for embeddings and GPT model access

​
Example

  • Input: “How does our refund policy work for international customers?”
  • Output: A clear, accurate response pulled directly from your uploaded refund policy documents.

​
Built With

  • LLMC Framework
  • Prompt Optimizer Flow
  • LLMC Executor
  • RAG Architecture (Vector Retrieval + Prompting)
  • OpenAI GPT Models
Deliver precise, document-grounded AI responses with automation, retrieval intelligence, and optimized prompting.