Langflow RAG Project: Citizen Advice Benefits Assistant

🎥 Watch Demo Video - See the system in action!

This project implements a Retrieval-Augmented Generation (RAG) system using Langflow to create an AI assistant that can answer questions about UK benefits using information from the Citizens Advice website.

🎯 Project Overview

The system scrapes content from Citizens Advice Benefits pages, processes it into embeddings, stores it in a Chroma vector database, and uses Mistral AI to generate contextual responses to user queries about UK benefits.

📋 Prerequisites

Before setting up this project, you'll need:

1. API Keys:

   • OpenAI API Key (for embeddings)
   • Mistral AI API Key (for text generation)

2. System Requirements:

   • macOS, Windows, or Linux
   • At least 4GB RAM (recommended)
   • Stable internet connection

🚀 Installation Steps

Step 1: Install Langflow Desktop (Recommended)

The easiest way to get started is using Langflow Desktop:

1. Download Langflow Desktop:

   • Visit Langflow Desktop
   • Click "Download Langflow"
   • Enter your contact information
   • Download the appropriate version for your OS

2. Install the Application:

   • macOS: Mount and install the downloaded .dmg file
   • Windows: Double-click the downloaded .msi file and follow the installation wizard
   • Linux: Follow the platform-specific installation instructions

3. Launch Langflow:

   • Open the Langflow application
   • The interface should be available at http://127.0.0.1:7860

Alternative: Install Langflow OSS (Python Package)

If you prefer the Python package version:

1. Install Python 3.10-3.13 (3.10-3.12 for Windows)

2. Install uv package manager:

   curl -LsSf https://astral.sh/uv/install.sh | sh

3. Create a virtual environment:

   uv venv langflow-env
   source langflow-env/bin/activate  # On Windows: langflow-env\Scripts\activate

4. Install Langflow:

   uv pip install langflow

5. Run Langflow:

   uv run langflow run

🔧 Project Setup

Step 2: Import the Flow

1. Open Langflow Interface:

   • Navigate to http://127.0.0.1:7860 in your browser
   • You should see the Langflow dashboard

2. Import the Flow:

   • Click on "My Projects" in the sidebar
   • Click the "Add Flow" button
   • Drag and drop the Citizen Advice RAG.json file from the langflow_json/ folder
   • The flow will be imported and appear in your projects list

Step 3: Configure API Keys

1. Access Global Variables:

   • In the Langflow interface, click on the settings icon (gear) in the top right
   • Select "Global Variables" or "Environment Variables"

2. Add Required API Keys:

   • OpenAI API Key:

     • Variable Name: OPENAI_API_KEY
     • Value: Your OpenAI API key from platform.openai.com

   • Mistral API Key:

     • Variable Name: MISTRAL_API_KEY
     • Value: Your Mistral AI API key from console.mistral.ai

3. Save Configuration:

   • Click "Save" to store your API keys
   • The keys will be securely stored and used by the flow components

📱 Sharing Your Prototype via SMS

While deploying this prototype to the internet with cloud services and Docker is out of scope for this demonstration, you can still share your working prototype with colleagues and end users for valuable feedback using SMS integration.

Why SMS Integration?

This approach allows you to:

• Get real user feedback without complex deployment
• Test with actual end users in a familiar medium (text messages)
• Validate your prototype before investing in full deployment
• Gather insights on user needs and pain points
• Demonstrate value to stakeholders with working examples

Setup Instructions

Follow the detailed instructions in the twilio/ folder:

1. Install Dependencies:

   pip install flask twilio python-dotenv requests

2. Set Up Twilio Account:

   • Sign up for a free Twilio account at twilio.com
   • Purchase a phone number (free trial available)
   • Get your Account SID and Auth Token

3. Configure Environment Variables: Create a .env file in the twilio/ directory with your Twilio credentials

4. Set Up ngrok for Local Development:

   • Install and authenticate ngrok
   • Use the provided run_ngrok.sh script for easy setup

5. Configure Twilio Webhook:

   • Point your Twilio phone number's webhook to your ngrok URL + /webhook

6. Run the Application:

   • Start your Langflow server (port 7860)
   • Start the Twilio webhook server (port 8888)
   • Start ngrok tunnel

How It Works

1. User sends SMS to your Twilio phone number
2. Twilio webhook receives the message
3. API request is sent to your Langflow flow
4. RAG system processes the question using Citizens Advice data
5. Response is sent back to the user via SMS

Example User Interactions

Users can ask questions like:

• "What is Universal Credit?"
• "How do I apply for Housing Benefit?"
• "Am I eligible for Personal Independence Payment?"
• "What benefits are available for disabled people?"

The system will respond with helpful, contextual information from the Citizens Advice knowledge base.

Benefits for Prototyping

• Low barrier to entry - Users just need to send a text message
• Real-world testing - Test with actual user scenarios
• Immediate feedback - Get responses and reactions quickly
• Cost-effective - Minimal setup costs compared to full deployment
• Scalable feedback - Easy to share with multiple stakeholders

For complete setup instructions, see the twilio/README.md file.

🎮 Using the RAG System

Step 4: Run the Flow

1. Open the Flow:

   • Click on the "Citizen Advice RAG" project in your projects list
   • The flow canvas will open showing the complete RAG pipeline

2. Initialize the System:

   • The flow includes components for:
     • URL Component: Scrapes content from Citizens Advice benefits pages
     • Text Splitting: Breaks content into manageable chunks
     • Embeddings: Converts text to vector representations using Mistral AI
     • Chroma Database: Stores and retrieves relevant information
     • Mistral Model: Generates responses based on retrieved context

3. Start the Playground:

   • Click on the "Playground" tab in the interface
   • This opens the chat interface where you can interact with the system

Step 5: Ask Questions

You can now ask questions about UK benefits, such as:

• "What is Universal Credit and how do I apply?"
• "Am I eligible for Housing Benefit?"
• "How does volunteering affect my benefits?"
• "What benefits are available for disabled people?"
• "How do I claim Child Benefit?"
• "What is Personal Independence Payment?"

The system will:

1. Search the Citizens Advice knowledge base
2. Retrieve relevant information
3. Generate contextual, accurate responses
4. Provide source information when available

🔍 Flow Architecture

The RAG system consists of several key components:

User Query → Chat Input → Chroma Search → Context Retrieval → Prompt Template → Mistral Model → Response

Key Components:

• URL Component: Scrapes Citizens Advice benefits pages
• SplitText: Breaks content into semantic chunks for better retrieval
• Mistral AI Embeddings: Creates vector representations of text
• Chroma Database: Stores embeddings and enables similarity search
• Prompt Template: Structures the context and question for the LLM
• Mistral Model: Generates human-like responses based on retrieved context

🛠️ Troubleshooting

Common Issues:

1. "No module named 'langflow.main'" Error:

   • Use uv run langflow run instead of langflow run
   • Reinstall with: uv pip install langflow -U --force-reinstall

2. API Key Errors:

   • Ensure API keys are correctly set in Global Variables
   • Verify keys are valid and have sufficient credits
   • Check for typos in variable names

3. Chroma Database Issues:

   • Clear cache if needed: Delete contents of ~/.cache/langflow/
   • Restart Langflow after clearing cache

4. Performance Issues:

   • Ensure you have at least 4GB RAM available
   • Close other resource-intensive applications
   • Consider using Langflow Desktop for better performance

💰 Costing

Understanding the costs involved in running this RAG system:

Langflow OSS (Local Setup)

• Cost: FREE 🎉
• Langflow open-source version has no licensing fees
• Runs entirely on your local machine
• No cloud hosting costs

Mistral AI API

• Free Tier: Very generous
• Embeddings: Free tier includes substantial usage
• Text Generation: Free tier available with reasonable limits
• Paid Tier: Pay-as-you-go pricing when you exceed free limits
• Recommendation: Start with free tier - it's sufficient for testing and moderate usage

OpenAI API

• Cost: Approximately $0.005 per query run
• Usage: Used for text embeddings in this flow
• Pricing: Based on token usage (very cost-effective for embeddings)
• Alternative: You can replace OpenAI embeddings with Mistral AI embeddings to reduce costs

Twilio SMS Integration

• Free Trial: 15,000 free SMS messages
• Paid Tier: ~$0.0075 per SMS (US/UK numbers)
• Phone Number: ~$1/month per number
• Recommendation: Free trial is sufficient for prototyping and initial user testing

Total Cost Estimate

• Setup: $0 (Langflow OSS is free)
• Per Query: ~$0.005 (mainly OpenAI embeddings)
• Per SMS: ~$0.0075 (Twilio)
• Monthly (100 queries via SMS): ~$1.25
• Monthly (1000 queries via SMS): ~$12.50

Cost Optimization Tips

1. Use Mistral AI for both embeddings and generation to eliminate OpenAI costs
2. Batch process queries when possible to reduce API calls
3. Monitor usage through your API provider dashboards
4. Set up usage alerts to avoid unexpected charges
5. Consider caching frequently asked questions to reduce API calls
6. Use Twilio free trial for initial prototyping and user testing

📚 Additional Resources

• Langflow Documentation
• Citizens Advice Benefits
• OpenAI API Documentation
• Mistral AI Documentation
• Technical RAG Documentation - Detailed explanation of how the RAG system works and how to build similar flows

🤝 Contributing

This project demonstrates how to build a RAG system using Langflow in low code and now code environments which we think would be great for the charity sector to prototype and test new ideas. Feel free to:

• Modify the flow for different use cases
• Add new data sources
• Improve the prompt templates
• Share your improvements

📄 License

This project is open source and available under the MIT License.

---

Note: This system is for educational and demonstration purposes. For official benefits advice, always consult with Citizens Advice or relevant government agencies.