BankRAG 🏦

A Retrieval-Augmented Generation (RAG) system built with Langflow, Weaviate, and transformers for banking customer support automation. Main purpose of project were try Langflow.

Features

• Vector Database: Weaviate for efficient similarity search
• Embeddings: Sentence Transformers (multi-qa-MiniLM-L6-cos-v1) for text vectorization
• RAG Pipeline: Complete question-answering system with document retrieval
• Containerized: Docker Compose setup for easy deployment
• Custom Components: Langflow components for Weaviate and transformers integration

Project Structure

BankRAG/
├── data/                           # Dataset and training data
│   ├── qa_dataset.csv             # Q&A dataset
│   ├── qa_dataset.txt             # Formatted Q&A text
│   ├── train_our.jsonl           # Training data
│   └── data_gen.ipynb            # Data generation notebook
├── docker/                        # Docker configuration
│   ├── docker-compose.yml        # Multi-service setup
│   ├── Dockerfile                # Langflow container
│   └── docker.env               # Environment variables
├── flows/                         # Langflow workflows
│   └── Complete flow.json        # Main RAG flow
├── langflow_components/           # Custom Langflow components
│   ├── transformers_embedding_model_component.py
│   └── weaviate_server_component.py
├── test/                          # Testing and examples
│   ├── test.ipynb               # API testing notebook
│   └── res.json                 # Sample responses
└── requirements.txt              # Python dependencies

Flow Structure

The Langflow workflow consists of interconnected components that handle document ingestion, embedding generation, vector storage, query processing, and response generation through a complete RAG pipeline.

Installation & Setup

Prerequisites

• Docker & Docker Compose
• Python 3.11+
• Git

Quick Start

1. Clone the repository

   git clone https://github.com/RailSAB/BankRAG.git
   cd BankRAG

2. Start all services

   cd docker
   docker-compose up -d

   There are can occur problems with langlow in container. If so: run langflow manually by:

   pip install -r requirements.txt
   langflow run

3. Access the interfaces

   • Langflow UI: http://localhost:7860
   • Weaviate: http://localhost:8080
   • Transformers API: http://localhost:8081

Services

Weaviate (Vector Database)

• Port: 8080
• Purpose: Stores document embeddings for similarity search
• Configuration: Text2vec-transformers vectorizer

Transformers Inference API

• Port: 8081
• Model: sentence-transformers-multi-qa-MiniLM-L6-cos-v1
• Purpose: Generates embeddings for text

Langflow

• Port: 7860
• Purpose: RAG workflow orchestration
• Credentials: admin/admin123

Usage

API Endpoint

The main RAG system is accessible via the Langflow API:

import requests
import uuid

url = "http://localhost:7860/api/v1/run/{FLOW_ID}"

payload = {
    "output_type": "chat",
    "input_type": "chat",
    "input_value": "Как посмотреть лимиты счета?",
    "session_id": str(uuid.uuid4())
}

headers = {
    "Content-Type": "application/json",
    "x-api-key": API_KEY
}

response = requests.post(url, json=payload, headers=headers)
result = response.json()
print(result)

Direct Embedding API

Test the transformers service directly:

curl -X POST "http://localhost:8081/vectors" \
  -H "Content-Type: application/json" \
  -d '{"text": "Sample text for embedding"}'

Further Development

Container and Configuration Issues

• Langflow Containerization Problems: Resolve issues with Langflow container deployment and ensure stable operation within Docker Compose environment
• Component Variable Configuration: Fix environment variable passing and component configuration issues in containerized Langflow setup
• Dependencies Management: Address package installation and module import problems in the container environment

API Integration Enhancement

• Langflow API Data Retrieval: Implement robust response parsing and error handling for Langflow API interactions
• Response Format Standardization: Create consistent data structures for API responses with proper error handling and validation
• Session Management: Improve session handling and conversation context management across API calls

Telegram Bot Integration

• Bot Framework Setup: Integrate the RAG system with Telegram Bot API for user interactions
• Message Processing Pipeline: Implement message queuing and processing for handling multiple concurrent users
• Rich Response Formatting: Add support for formatted messages, buttons, and interactive elements in Telegram responses
• User Session Management: Implement user-specific conversation contexts and history tracking