QRAS Semantic Memory

What it does

The Qdrant RAG System is a comprehensive Retrieval-Augmented Generation (RAG) system that combines semantic search with LLM generation. It provides both command-line interface (CLI) tools and a web interface for indexing documents, performing searches, and engaging in context-aware chat powered by Large Language Models (LLMs). The system supports multiple data sources and is designed to produce token-efficient output optimized for AI agents.

When to use it

• Knowledge Base Search: Quickly find relevant information within a collection of documents when needing answers or insights. For example, searching internal documentation for specific procedures.
• Chatbot Development: Build chatbots that can answer questions based on your own data, providing more accurate and contextually relevant responses than relying solely on the LLM's pre-existing knowledge.
• Research & Analysis: Efficiently explore large datasets of text documents to identify patterns, trends, or specific information.
• Internal Tooling: Create a searchable interface for internal teams to access company policies, training materials, or other critical resources.

Key capabilities

• CLI Interface for indexing and searching
• Semantic Vector Search
• RAG Chat System with LLM integration
• Support for JSON files and markdown directories as data sources
• Hybrid Search (vector similarity + keyword matching)
• Incremental Indexing of documents
• LLM-Optimized Output format
• Interactive search and chat interfaces
• Streaming responses from the LLM
• Web Interface with FastAPI backend and Svelte frontend

Example prompts

• ./qras query "machine learning algorithms" - Performs a semantic search for information on machine learning algorithms.
• ./qras chat "What is neural network?" --chat-model llama3 – Initiates an interactive chat session using the Llama3 model to answer questions about neural networks.
• ./qras index --input-path ./documents --collection docs - Indexes all files within the 'documents' directory into a collection named 'docs'.

Tips & gotchas

• Prerequisites: Requires Python 3.9+, Ollama, and Qdrant to be installed and running before use. Ensure these dependencies are properly configured prior to attempting any indexing or searching operations.