ZRSVN RAG

Web application with lexical, semantic, and hybrid search capabilities and answer generation based on selected documents.

Features

• Interactive Leaflet map for drawing bounding boxes
• Three search modes: lexical (BM25), semantic (pgvector), and hybrid (RRF)
• Conversational interface via Azure OpenAI (default model is GPT-4o-mini)
• Generation of presigned URLs for viewing contextual PDF documents
• User feedback logging
• Observability via Logfire integration

Technical requirements

• Python 3.8+
• PostgreSQL with the pgvector and ParadeDB extensions
• MinIO/S3 storage for PDF documents
• Azure OpenAI API access

Installation

1. Clone the repository:

   git clone https://github.com/gregorgatej/zrsvn-rag.git
   cd zrsvn-rag

2. Install dependencies:

   pip install -r requirements.txt

3. Create a .env file with the following variables:

   POSTGRES_PASSWORD=your_postgres_password
   S3_ACCESS_KEY=your_s3_access_key
   S3_SECRET_ACCESS_KEY=your_s3_secret_key
   ZRSVN_AZURE_OPENAI_ENDPOINT=your_azure_endpoint
   ZRSVN_AZURE_OPENAI_KEY=your_azure_openai_key

4. Prepare the PostgreSQL database:

• Create the zrsvn database with the rag_najdbe schema
• Install the pgvector and paradedb extensions

Running the application

python app.py