AI Knowledge Base (RAG Service) 🧠📚🔍

An AI microservice that implements the Retrieval-Augmented Generation (RAG) architecture. This system enables a Large Language Model (LLM) to answer questions based on a specific, private knowledge base, eliminating hallucinations and providing factual, context-aware responses.

🧠 Core Concept: Giving an LLM a Long-Term Memory

The RAG architecture solves a fundamental problem with LLMs: their knowledge is static and generic. RAG connects the LLM to an external source of truth through two main phases:

1. Indexing Pipeline (Offline): An ETL process where documents (text, PDFs, etc.) are loaded, split into smaller "chunks," and each chunk is converted into a numerical vector (an "embedding") using a specialized model. These vectors, representing the semantic meaning of the text, are stored in a Vector Database.
2. RAG Service (Online): When a user asks a question, the system:
   • Retrieves: Converts the question into a vector and uses it to search the database for the most relevant text chunks (similarity search).
   • Augments: Inserts these retrieved chunks into a prompt along with the original question, instructing the LLM to use only this information to answer.
   • Generates: Sends the augmented prompt to the LLM, which generates a factual answer grounded in the provided context.

🚀 Engineering & MLOps Highlights

This project demonstrates the end-to-end construction of a production-ready Question & Answering system.

• Offline Indexing Pipeline: A modular and reusable script (indexer.py) that processes and vectorizes knowledge using LangChain (for text splitting), SentenceTransformers (for embeddings), and ChromaDB (for vector storage).
• Optimized RAG Inference Service: A FastAPI service that loads the models and database once on startup to ensure low-latency responses.
• Hallucination Reduction: The augmented prompt is designed to instruct the LLM to be factual and strictly base its answers on the provided context, a critical technique for business applications.
• Standardized Infrastructure: The application is fully containerized with Docker (including the vector database) and validated by a GitHub Actions pipeline, proving the robustness of our "AI service factory" pattern.

🏗️ System Architecture

This diagram illustrates the two phases of the RAG system:

graph TD
    subgraph "Phase 1: Indexing Pipeline (Offline)"
        A[Documents .txt] --> B{Text Splitter};
        B --> |Chunks| C{Embedding Model};
        C --> |Vectors| D[(Chroma Vector DB)];
    end

    subgraph "Phase 2: RAG Service (Online)"
        E[User] -->|Question| F(FastAPI Service);
        F -->|Embed Question| G{Embedding Model};
        G -->|Similarity Search| D;
        D -->|Relevant Context| F;
        F -->|Build Augmented Prompt| H[LLM];
        H -->|Factual Answer| F;
        F -->|JSON Response| E;
    end

Loading

🏁 Getting Started

Prerequisites

• Git
• Python 3.9+
• Docker Desktop (running)
• An OpenAI API Key

1. Preparing the Knowledge Base (Indexing)

This step needs to be run only once, or whenever the knowledge.txt file is updated.

1. Clone the repository:

   git clone https://github.com/PRYSKAS/ai_knowledge_base.git
   cd ai_knowledge_base

2. Install dependencies:

   pip install -r requirements.txt

3. (Optional) Customize Knowledge: Edit the knowledge.txt file in the project root with your own text content.
4. Run the Indexing Pipeline:

   python -m indexing_pipeline.indexer

   This will create (or update) the chroma_db folder containing the vector embeddings of your knowledge base.

2. Running the Q&A Service

1. Set up environment variables:

   • Create a .env file from the example: copy .env.example .env (on Windows) or cp .env.example .env (on Unix/macOS).
   • Add your OPENAI_API_KEY to the new .env file.

2. Run locally using Uvicorn (for development):

   uvicorn serving.main:app --reload --port 8001

   Access the API documentation and interact with the service at http://127.0.0.1:8001/docs.

3. Run using Docker (Recommended for stable execution):

   • Build the Docker image:

     docker build -t ai-knowledge-base-service .

     (This copies the chroma_db folder into the image)
   • Run the container:

     docker run -d -p 8001:8001 --env-file .env --name ai-kb-rag ai-knowledge-base-service

   Access the API at http://127.0.0.1:8001/docs.

---