ChromaQuery: AI-Powered Knowledge Retrieval

ChromaQuery is an AI-powered knowledge retrieval system that integrates retrieval-augmented generation (RAG), web scraping, and ChromaDB for accurate and real-time responses. The system uses OpenAI embeddings and vector search to retrieve relevant articles and generate contextual answers.

The project consists of three parts:

• RAG (Retrieval-Augmented Generation): Combines document retrieval and language generation to enhance response accuracy using OpenAI embeddings and vector-based search.
• Web Scraping: Extracts and updates content from online sources using BeautifulSoup and requests.
• ChromaDB: Stores and indexes documents and their embeddings for efficient retrieval using semantic search.

RAG Pipeline Architecture

Project Architecture

The ChromaQuery system follows a modular design that integrates several key components for efficient knowledge retrieval and response generation. Below is an overview of the main architecture:

1️⃣ User Input & Query Processing

• User Input: The system starts by accepting a query from the user (e.g., "Tell me something related to crypto").
• Query Embedding: The input query is converted into an embedding using OpenAI's model, which transforms the text into a numerical vector.

2️⃣ Document Retrieval with ChromaDB

• Article Fetching: The system fetches technology-related articles from a predefined source using web scraping (via BeautifulSoup and requests).
• Metadata & Embedding Storage: For each article, the system stores the title, link, and embedding (semantic vector) in ChromaDB, along with metadata for easy retrieval.
• Persistence: ChromaDB provides persistent storage to keep document embeddings and metadata across sessions.

3️⃣ Relevant Document Search

• Query Matching: Once the user query is embedded, it is matched against the embeddings stored in ChromaDB using vector-based search to find the most relevant documents.
• Top Results: The system retrieves the top documents based on semantic similarity to the query embedding.

4️⃣ Web Scraping for Article Content

• Full Text Retrieval: After retrieving relevant article links, the system scrapes the content of these articles (handling multiple sources) to extract detailed information.
• Text Chunking: The article content is split into smaller chunks for more granular search and indexing.

5️⃣ Storing Chunks in ChromaDB

• Chunk Storage: The system stores these article chunks (along with their embeddings) in a new ChromaDB collection, ensuring fast access and retrieval.

6️⃣ Final Response Generation

• Answer Synthesis: The system then queries the ChromaDB collection to retrieve the most relevant chunks and generates a contextual response using the OpenAI model.

Diagram of the Architecture

Get Article & Link

graph LR
    A[User Input] --> B[Generate Query Embedding]
    B --> C[ChromaDB - Store Embeddings and Metadata]

Loading

Get Article Content

graph LR
    C[ChromaDB - Store Embeddings and Metadata] --> D[Web Scraping of Articles]
    D --> E[Store Article Chunks in ChromaDB]

Loading

Retrieve Chunks & Generate Response

graph LR
    E[Store Article Chunks in ChromaDB] --> F[Retrieve Relevant Chunks]
    F --> G[Generate AI-Generated Response]

Loading