FastAPI RAG with PostgreSQL and Nebius

Overview

This project demonstrates a Retrieval-Augmented Generation (RAG) implementation using FastAPI as the web framework, PostgreSQL (managed by Nebius) as the database, and vector search for chunk similarity. It processes uploaded files (text or PDF), extracts content, embeds the content using vector embeddings, and provides endpoints to ask questions based on the embedded content.

Features

• File Upload: Upload text (.txt) or PDF (.pdf) files to the server.
• Content Extraction: Extracts text content from files, with OCR fallback for PDFs without readable text.
• Database Storage: Saves extracted content and file metadata in PostgreSQL.
• Vector Embedding: Uses vector embeddings for content similarity search.
• Ask Questions: Query the database for context-aware answers.
• API Endpoints: Includes endpoints for uploading files, querying context, and retrieving similar chunks.

---

Tech Stack

• Backend: FastAPI
• Database: PostgreSQL (Managed by Nebius)
• Vector Embedding: pgvector for vector similarity search.
• PDF/Text Parsing:
  • PyPDF2 for PDF parsing.
  • pytesseract for OCR.
• Environment Management: dotenv
• Embedding Model: Integrated with vector models like BAAI/bge-en-icl.

---

Prerequisites

• Python 3.8+
• PostgreSQL with pgvector extension enabled.
• Installed Tesseract OCR (for OCR capabilities).
• .env file for configuration.

Example .env File:

POSTGRES_USERNAME=your_postgres_username
POSTGRES_PASSWORD=your_postgres_password
POSTGRES_HOST=your_postgres_host
POSTGRES_PORT=5432
DATABASE_NAME=fastapi_rag_noframework_db

NEBIUS_API_KEY=your_nebius_api_key

---

Setup Instructions

1. Clone the Repository

git clone <repository-url>
cd <repository-folder>

2. Set Up a Virtual Environment

python3 -m venv venv
source venv/bin/activate

3. Install Dependencies

pip install -r requirements.txt

4. Configure PostgreSQL

Ensure PostgreSQL is running and pgvector extension is installed. The database will be created automatically if it does not exist.

---

Usage

1. Start the Server

Run the FastAPI app:

uvicorn main:app --reload

2. API Endpoints

Endpoint	Method	Description
/	GET	Lists all uploaded files.
/uploadfile/	POST	Upload a file (text or PDF).
/ask/	POST	Ask a question using the document's context.
/find-similar-chunks/{file_id}	POST	Retrieve similar content chunks for a file and question.

Example Request: Upload File

curl -X 'POST' 'http://localhost:8000/uploadfile/' -F "file=@sources/obama.txt"

Example Request: Ask Question

curl -X 'POST' 'http://localhost:8000/ask/' \
-H "Content-Type: application/json" \
-d '{
  "question": "Who is the president of the United States according to the context provided?",
  "document_id": 1
}'

---

File Parsing

Supported File Types

• .txt
• .pdf (with OCR fallback for scanned PDFs)

Parser Workflow

1. Text Extraction:
   • Plain text files: Simple file reading.
   • PDFs: Uses PyPDF2 for text-based PDFs and pytesseract for OCR.
2. Repair and Validation:
   • Repairs malformed PDFs using pikepdf.
   • Validates PDF structure before parsing.

---

Database

Tables

1. files:
   • Stores file metadata and content.
   • Columns: file_id, file_name, file_content.
2. file_chunks:
   • Stores content chunks with vector embeddings.
   • Columns: chunk_id, file_id, chunk_text, embedding_vector.

Deleting Tables

The project includes a utility to drop all tables:

python delete_tables.py

---

Testing

The project includes automated test scripts:

python tests/test_parsers.py

These tests are not written with best practices in mind but were just create to quickly see if the main functionalities are working.

---

Future Enhancements

• Add support for additional file types (e.g., Word documents).
• Implement authentication for secure API usage.
• Enhance chunking logic for better context extraction.

---

Contributing

Contributions are welcome! Please fork the repository and submit a pull request.

---

License

This project is licensed under the MIT License. See the LICENSE file for details.