---

RAG Chatbot Project

A step-by-step overview of a Retrieve-and-Generate (RAG) chatbot using PostgreSQL + pgvector, Ollama LLM, and a Tkinter GUI.

---

Requirements

Before starting, make sure you have the following installed:

• PostgreSQL (v14+ recommended)
• Ollama (for running a local LLM)
• Python 3.10+ and the following Python packages:
  • psycopg2
  • numpy
  • sentence-transformers
  • (tkinter is usually bundled with Python)

You can install the required Python packages with:

pip install psycopg2 numpy sentence-transformers

Note:

• requests and tqdm are not required for this project based on the current codebase.
• Make sure PostgreSQL server is running and pgvector is properly installed before proceeding.

---

1. Overview

• Store document embeddings in PostgreSQL with pgvector
• Retrieve top-k matches via vector similarity search
• Generate answers using a local Ollama LLM
• Interact through a simple Tkinter-based chat interface

2. Install pgvector

• Docker: Pull and run the pgvector/pgvector image (bundles PostgreSQL and pgvector).
• Manual: Clone the pgvector GitHub repository and run:

  make
  make install

  to build and install the extension.

3. Enable Extension and Define Schema

• Enable the extension inside your database:

  CREATE EXTENSION IF NOT EXISTS vector;

• Create a documents table:

  CREATE TABLE documents (
    id SERIAL PRIMARY KEY,
    content TEXT,
    embedding VECTOR(<dimension>)  -- e.g., VECTOR(1024)
  );

4. RAG Workflow

1. Ingest: Encode documents into 1024-dimensional float vectors and store them in the documents table.
2. Query: Encode a user's question into a query vector.
3. Search: Perform a nearest-neighbor search using SQL operators (<->, <=>, <#>, <+>) to find the top-k matching documents.
4. Generate: Concatenate retrieved snippets and pass them to Ollama to generate a context-grounded answer.

5. Project Modules

• add.py: Insert text documents and their embeddings into the database.
• brain.py: Core RAG logic — retrieve relevant documents (query_postgresql) and generate responses (generate_response).
• delete.py: Remove documents by ID or custom criteria.
• create_table.py: Initialize the documents table schema.
• config.py: Load environment variables and initialize database connections and the embedder model.
• chatbot.py: Tkinter-based GUI for sending user queries and displaying bot responses.

6. Key Concepts

• Vector Storage: pgvector enables efficient storage and indexing of high-dimensional arrays.
• Similarity Search: SQL operators calculate distances (Euclidean, cosine, inner-product, or Manhattan) between vectors.
• RAG Pattern: Retrieve relevant context → Generate an informed, grounded answer.

7. Getting Started

• Populate your vector database using add.py.
• Launch the Tkinter chat client with chatbot.py.
• Ask questions and receive contextually accurate responses in real time.

---

This project demonstrates a clean, modular RAG architecture using fully local components — ideal for building private, efficient, and extensible AI chatbots.

---