Mastering Agentic AI with Langflow and Astra DB 🚀

Workshop Overview 🛠️✨

Welcome to this transformative workshop! Today, we’ll explore how to build an Agentic AI Application using Langflow and Astra DB. By the end of this workshop, you’ll have created a powerful customer support system capable of:

Handling FAQs with contextual accuracy.
Processing customer orders effortlessly.
Retrieving real-time product details through seamless integrations.

Key Technologies

Langflow: A low-code platform enabling drag-and-drop simplicity for AI workflows.
Astra DB: A vector-enabled database designed for real-time semantic search and advanced querying.

What You’ll Learn:

How to implement retrieval-augmented generation (RAG) for precise FAQ handling.
Building multi-agent systems to tackle complex queries.
Deploying an AI-powered customer support agent with a Streamlit front end.

Get ready to build scalable, real-world AI applications that make a difference.

Use Case: Empowering Customer Support with AI 🤝💡

Why This Use Case Matters

E-commerce businesses face constant customer inquiries about:

Shipping times
Order statuses
Product details

Challenges:

Manual responses are time-consuming and prone to errors.
Scaling support teams to meet demand is costly.

Solution: An AI-driven customer support system can:

Efficiently answer FAQs using retrieval-augmented generation.
Access real-time order and product data from a centralized database.
Scale effortlessly to handle high query volumes with minimal intervention.

Workshop Goal 🎯

By the end of this workshop, you’ll build a system capable of:

Answering FAQs with RAG.
Querying Orders using Astra DB.
Collaborating Agents to address multi-faceted queries.

Let’s dive in!

Prerequisites ✅

This workshop assumes you have access to:

A GitHub account.
A paid OpenAI account. (We will provide API keys if needed.)

Additionally, create free accounts for:

Streamlit: Follow the quickstart guide.
Langflow: Sign up here.

Let’s Make Magic Happen! 🦄✨

We’ll begin by implementing a multi-agent-based flow for our frontend application. This flow will be the backbone for delivering accurate and dynamic responses to customer inquiries.

Step 1: Set Up Our Flow 🛠️🌟

Start by setting up a new flow in Langflow, the foundation for your project.

Create a New Flow ➕

Open Langflow and click New Flow on the dashboard.
This will open the flow creation interface.

Choose a Blank Flow Template 📄

Select the Blank Flow option from the available templates.
Langflow offers a variety of templates for different AI workflows, which are continuously expanding.

Name Your Flow ✏️

Click Flow Settings to configure the flow details.
Provide a meaningful name that reflects the flow's purpose.
Define the API endpoint to make the flow accessible via external HTTP calls.

🎉 Success! Your flow is now set up and ready for development. Next, create a simple echo flow to see how components interact.

Step 2: Create a Simple Echo Flow 🔁💬

Learn the basics of Langflow by creating a simple flow where the input is echoed back to the user. This exercise introduces the interface and demonstrates how components interact.

Add Components 🧩

Drag Chat Input and Chat Output components from the left-hand panel onto the canvas.
These components represent the user input and the system response.

Connect the Components 🔗

Link the Chat Input component to the Chat Output component.
This creates a direct flow, sending user input directly to the output.

Test in the Playground 🎮

Click the Playground button to test the flow.
In the Playground, type:
```
hello
```
Observe how the input is echoed back in the output.

🎉 Success! You’ve created your first flow in Langflow. This simple echo flow demonstrates how components interact. Next, extend this flow by introducing an agent to handle more complex tasks.

Step 3: Extend the Flow with an Agent 🤖💡

Extend your flow by integrating an agent to enable advanced capabilities. The agent will act as the brain behind your application, processing inputs and interacting with tools.

Remove Existing Connections ✂️

Disconnect the Chat Input and Chat Output components.

Add an Agent Component 🧩

Drag and drop the Agent component onto the canvas.

Connect the Components 🔗

Link the Chat Input component to the Agent component.
Connect the Agent component to the Chat Output component.

Configure API Credentials 🔑

The agent supports various LLM providers. For this workshop, we use OpenAI, which requires an API key.
Click the icon in the OpenAI API Key field of the agent.
Add a new variable of type Credential, name it OPENAI_API_KEY, and select the variable.

Set Agent Instructions 📜

Instructions guide the agent on what tasks to perform and how to interact with tools.

Click the icon in the Agent Instructions field to view the default instruction:

You are a helpful assistant that can use tools to answer questions and perform tasks.

For now, leave the default instruction unchanged.

Test the Flow in the Playground 🎮

Open the Playground and test the agent by entering a query like:
```
Who is Bob Marley?
```
Observe the agent's response and verify that it processes the input correctly.

🎉 Success! You’ve successfully integrated an agent into your flow. This is a key step toward creating a smart and interactive application. Next, we’ll tailor the agent for our specific use case: the "Customer Support Agent."

Step 4: Implement the RAG Flow 🔍📚

The agent we developed so far cannot answer questions specific to our context and may hallucinate, producing irrelevant or incorrect responses. To address this, we’ll implement a Retrieval-Augmented Generation (RAG) flow. This approach allows the agent to retrieve relevant data from the company’s FAQ database for accurate responses.

Add an Astra DB Component 🗄️

Drag an Astra DB component onto the canvas from the Vector Stores section.
Switch the Astra DB component into Tool Mode
Use this tool to retrieve vectorized FAQ data.
This tool enables the agent to search the FAQ database for relevant information.

Connect the Components 🔗

Link the Astra DB tool to the Agent component.

Create a New Database in Astra DB 📋

In the Astra DB component, click Add New Database to create a new database.
Fill in the required details and click Create Database.

Note: Database creation may take a few minutes.

Switch to Astra DB 🔄

Open Astra DB and navigate to your newly created database.

Create an FAQ Collection 🗂️

In Astra DB, go to Data Explorer and click Create Collection.
Name the collection faq, enable the Vector-Enabled Collection option, and click Create Collection.

Switch Back to Langflow ↩️

Return to Langflow to continue configuring the RAG flow.

Configure the Astra DB Component ⚙️

Fill in the following details in the Astra DB component:
- Collection Name: faq

Extend the tool description. Click Edit tools and extend the Tool Description with:

Answer frequently asked questions (FAQs) about shipping, returns, placing orders, and more.

Define Agent Instructions ✍️

Add the following instructions to the Agent Instructions field:

Your primary responsibility is to use the available tools to accurately address user inquiries and provide detailed, helpful responses. You can:

- Answer frequently asked questions (FAQs) about shipping, returns, placing orders, and more.

Example: If the ask is about delivery times, check the FAQ.
Always aim to deliver clear, concise, and user-focused solutions to ensure the best possible experience.

Test the RAG Flow 🎮

Open the Playground and test the flow with a query like:
```
What are the shipping times?
```
Observe the response to confirm the agent is unable to respond meaningfully.

Verify the Flow ✅

Test Point 1: Observe that the agent is unable to respond meaningfully without context.

🎉 Success! Your agent now includes a RAG flow to answer context-specific queries accurately. The foundation is laid—next, we’ll enrich the agent with the required FAQ data. Say goodbye to hallucinations and hello to reliable responses!

Step 5: Vectorize the FAQ 📚✨

The agent cannot yet provide proper answers because it lacks the necessary information from the FAQ. Although we have an FAQ collection in the database, it is currently empty. Let’s populate it using a vectorization flow in Langflow.

Add Components to the Canvas 🧩

File Component:
- Drag a File component from the Data section.
- This will allow you to upload the Company_FAQ.pdf.
Split Text Component:
- Drag a Split Text component from the Processing section.
- This will divide the FAQ into manageable chunks for vectorization.
Astra DB Component:
- Drag an Astra DB component from the Vector Stores section.
- This will store the vectorized data.

Connect the Components 🔗

Link the File component’s Data endpoint to the Split Text component’s Data Inputs endpoint.
Connect the Chunks output of the Split Text component to the Ingest Data endpoint of the Astra DB component.

Upload the FAQ Document 📥

Upload the Company_FAQ.pdf file via the File component.

Configure the Astra DB Component ⚙️

Fill in the following details:
- Database Name: customer-support
- Collection Name: faq

Run the Vectorization Flow ▶️

Click the play button on the Astra DB component to execute the flow.
The PDF will be split into chunks, vectorized, and stored in Astra DB alongside their corresponding vectors.

Test in the Playground 🎮

Open the Playground and query the system with:
```
What are the shipping times?
```
Confirm that the system retrieves accurate responses based on the newly vectorized FAQ data.

Verify Vectorization ✅

Test Point 1: Ensure the faq collection in Astra DB contains the vectorized data by querying it directly in Astra DB.
Test Point 2: Use queries like "What are the shipping times?" in the Langflow Playground to confirm the agent retrieves accurate information from the FAQ.

🎉 Success! Your FAQ has been successfully vectorized and integrated. The agent is now equipped to handle context-specific queries with precision.

Step 6: Upload Sample Data 📦💾

Enable your customer support system to retrieve order and product details by uploading sample data to Astra DB.

Switch to Astra DB 🔄

Navigate from Langflow to Astra DB.

Create the `Orders` Collection 📝

In Astra DB, open your customer-support database and go to the Data Explorer.
Click Create Collection and configure:
- Collection Name: orders
- Disable the Vector-enabled collection switch.
- Click Create Collection.

Create the `Products` Collection 🛒

Repeat the steps above to create another collection:
- Collection Name: products
- Disable the Vector-enabled collection switch.
- Click Create Collection.

Load Data into the Collections 📥

Orders Collection:
- Select the orders collection in the Data Explorer.
- Click Load Data and upload the file: sample_orders.csv.
Products Collection:
- Select the products collection in the Data Explorer.
- Click Load Data and upload the file: sample_products.csv.

Verify Data Uploads ✅

Test Point 1: Query the orders collection in Astra DB to ensure the data was uploaded correctly.
Test Point 2: Verify the products collection reflects the sample data.

Return to Langflow 🔙

After completing the data uploads, switch back to Langflow to continue building your flows.

🎉 Success! Your sample data is now loaded into Astra DB, ready for integration with the OrderLookupAgent and other flows.

Step 7: Build the Order Lookup Agent Flow 🛠️🔍

Create a flow that retrieves order details and related product information from the database, enabling your customer support system to address queries effectively.

Add Astra DB Components 📦

Drag and drop an Astra DB component for the orders collection:
- Tool Name: OrderLookup
- Database Name: customer-support
- Collection Name: orders
- Tool Description: A tool used to look up an order based on its ID.
- Tool Params: !orderNumber
Drag and drop another Astra DB component for the products collection:
- Tool Name: ProductLookup
- Database Name: customer-support
- Collection Name: products
- Tool Description: A tool used to look up a product based on its ID.
- Tool Params: !productId

Add an Agent Component 🤖

Drag and drop an Agent component to act as the logic handler.

Configure the agent:

OpenAI API Key: Add your key.

Agent Instructions:

You are an expert in analyzing customer orders and providing detailed and accurate information. Your primary role is to utilize the provided tools to efficiently look up order numbers, retrieve relevant details about the orders, and address any questions or concerns the user may have.

Lookup order numbers and product IDs using the tools provided.

Orders always contain an array of product IDs. Use these IDs to look up the specific products from the product lookup tool and aggregate the product information with the order to provide a clear summary of the order.

If the order does not exist, simply tell the user to try again as the ID wasn't found.

Only return information about orders; do not return anything else.

Enable Tool Mode and rename the agent to OrderLookupAgent.

Connect Components 🔗

Connect the Tool endpoints of the OrderLookup and ProductLookup components to the Tools endpoint of the OrderLookupAgent.

Validate the Flow ✅

Click the Play button on the OrderLookupAgent to verify the flow is configured correctly.

🎉 Congratulations! You’ve successfully built an OrderLookupAgent that retrieves and combines order and product data from your Astra DB collections. This forms a vital component of your customer support system!

Step 8: Combine Both Flows with a Manager 🛠️🤖

Integrate the RAG Flow and Order/Product Lookup Flow using a ManagementAgent. This agent will intelligently route user inquiries, leveraging both flows for accurate and complete responses.

Rename the RAG Agent ✏️

Rename the existing RAG agent to ManagementAgent to reflect its new role as a central coordinator.

Define ManagementAgent’s Role 📜

Provide the ManagementAgent with detailed instructions to guide its use of tools:

You are a skilled customer service manager and information router. Your primary responsibility is to use the available tools to accurately address user inquiries and provide detailed, helpful responses. You can:

- Look up order numbers to retrieve and share order details.
- Access product information to provide relevant descriptions or specifications.
- Answer frequently asked questions (FAQs) about shipping, returns, placing orders, and more.

If a query requires multiple tools, combine their outputs to deliver a comprehensive response.  
Example: For an inquiry about canceling an order, retrieve the order and product details, and also reference the FAQ for the cancellation policy.

Always aim to deliver clear, concise, and user-focused solutions to ensure the best possible experience.

Connect Tools to ManagementAgent 🔗

Link the OrderLookupAgent component’s Toolset endpoint to the ManagementAgent component’s Tools endpoint.

Test in the Playground 🎮

Click the Playground button and run this query:
```
What is the status of order #1001?
```
Verify the response combines relevant data from the Orders and Products collections.

Conduct Comprehensive Testing ✅

Test Point 1: Ensure responses match the data in the Orders and Products collections. 📋
Test Point 2: Run mixed queries to validate task routing, such as:
```
How can I cancel order #1001 and what is the shipping policy?
```
Confirm the agent combines multiple tools to provide aggregated responses. 🔄

🎉 Congratulations! Your ManagementAgent now seamlessly integrates the RAG and Order Lookup flows, enabling your system to handle complex, multi-faceted queries with ease.

Step 9: Integrate with a Python Front End 🖥️🐍

Create a user-friendly interface using Streamlit to connect your Langflow-powered backend with end users seamlessly.

Fork the Repository 🍴

Visit Agentic AI Workshop.
Click Fork in the top-right corner of the GitHub page.

Clone the Repository 📁

Clone the forked repository to your local machine:

git clone https://github.com/YOUR_GITHUB_USERNAME/agentic-ai-workshop.git
cd agentic-ai-workshop

Set Up a Virtual Environment 🐍

Create and activate a Python virtual environment:
```
python3 -m venv venv
```
- On Linux/Mac:
```
source venv/bin/activate
```
- On Windows:
```
venv\Scripts\activate.bat
```

Install Dependencies 📦

Install the required Python packages:
```
pip install -r ./requirements.txt
```

Configure Secrets 🔑

Copy and edit the example secrets file:

cp .streamlit/secrets.toml.example .streamlit/secrets.toml

Retrieve Langflow Credentials 🔐

Log in to Langflow, click API.
Click Generate Token.
Click the copy icon on the right-hand side. Paste the token into your secrets.toml.
Copy your Langflow ID from the curl command and paste it into your secrets.toml.
Click Flow Settings to see the Endpoint Name you have configured for the flow at the start of the workshop.
Copy the Endpoint Name and paste it into your secrets.toml.

Your secrets.toml file should now contain your credentials:

APP_TOKEN="AstraCS:LKeBuZvesCUlokSbiNfjCvAG:e291f0b1f37925cb31565d859bc56ec25cc1371..."
LANGFLOW_ID="cc011911-e624-4ec4-81d0-f1894f2..."
ENDPOINT="customer-support"

Run the Application ▶️

Start the Streamlit server:
```
streamlit run app.py
```

Test the Front End 💬

Open the app in your browser (http://localhost:8501).
Try sample questions or queries like:
- "How can I track my order?"
- "What is your return policy?"
- "Can I cancel order #1004?"

Understand How `app.py` Works 📝

The app.py file connects the Streamlit front end with the Langflow backend.
Key features include:

API Integration: The run_flow() function sends queries to Langflow and retrieves responses.
User Interface: Designed with Streamlit to offer a simple and intuitive experience.

For a detailed walkthrough of app.py, see Understanding app.py.

🎉 Success! Your Streamlit front end is up and running, offering an intuitive and professional way for users to interact with your AI-powered customer support system.

Step 10: Deploy the Application to Streamlit Cloud 🌐☁️

Now that your application is running locally, let’s deploy it to Streamlit Cloud for easy access and sharing. Streamlit Cloud allows you to host your application directly from your GitHub repository.

Prepare Your Repository 📁

Push any local changes to your forked repository:

git add .
git commit -m "Prepare for Streamlit Cloud deployment"
git push origin main

Log In to Streamlit Cloud 🌐

Visit Streamlit Cloud and log in using your GitHub credentials.

Create a New Deployment 🚀

On the Streamlit Cloud dashboard, click Create App.
Fill in the details, specifying your forked GitHub repository. Choose the branch (e.g., main) and app.py as the entry point.

Set Up Secrets 🔑

Go to the Advanced Settings section and locate the Secrets tab.

Copy the contents of your local secrets.toml file and paste them into the Secrets Editor. Include:

LANGFLOW_ID = "Your_Langflow_ID"
ENDPOINT = "Your_Langflow_Endpoint"
APP_TOKEN = "Your_Application_Token"

Deploy the Application ▶️

Click Deploy and wait for the build process to complete. Streamlit will notify you when the app is live.

Access and Test Your Application 🌍

Your app will be accessible via a unique URL (e.g., https://agentic-ai-workshop.streamlit.app).
Open the app in your browser and test it by:
- Trying sample questions like:
  - "How can I track my order?"
  - "What is your return policy?"
- Using custom queries such as:
  - "Can I cancel order #1001 and what is the shipping policy?"

Manage Your Deployment 🔄

Use the Streamlit Cloud dashboard to:
- Monitor app usage.
- Update the app by pushing changes to your GitHub repository.
- Redeploy as needed.

🎉 Success! Your AI-powered application is now live on Streamlit Cloud, ready to deliver real-time customer support solutions to users anywhere in the world.

Resources 📚🔗

📖 Langflow Documentation: Langflow Docs
🛠️ Astra DB Documentation: Astra DB Docs
💾 GitHub Repository: Workshop Content

Call to Action 🚀✨

🎉 Congratulations on building a cutting-edge AI system! Expand your skills further by exploring new use cases like inventory management, knowledge retrieval, or personalized recommendations. Let's innovate! 🚀

🤔 What will you build next? The tools are in your hands. 💡🌟

Appendix: Understanding `app.py` 📝

The app.py file serves as the backbone of your application, connecting the Streamlit front end with the Langflow-powered backend. Here's a breakdown of its key components:

API Integration

The run_flow() function is responsible for communicating with the Langflow backend:

Base API URL: Retrieved from the secrets.toml file (LANGFLOW_ID, ENDPOINT, and APP_TOKEN).
Payload: Sends user input to the backend in JSON format.
Response Handling: Extracts the AI-generated response for display in the UI.

Code Snippet:

def run_flow(message: str) -> dict:
    api_url = f"{BASE_API_URL}/lf/{LANGFLOW_ID}/api/v1/run/{ENDPOINT}"

    payload = {
        "input_value": message,
        "output_type": "chat",
        "input_type": "chat",
    }

    headers = {"Authorization": "Bearer " + APPLICATION_TOKEN, "Content-Type": "application/json"}
    response = requests.post(api_url, json=payload, headers=headers)
    return response.json()

Streamlit Front End

The Streamlit front end provides an intuitive interface for user interactions:

Chat Input and Output: Users submit queries, and AI responses are displayed as styled chat bubbles.
Sample Questions: Buttons provide quick access to predefined queries.
Custom Styling: CSS enhances the visual appeal of the chat interface.

Code Snippet (Input and Response Handling):

# Display chat history
for message in st.session_state["messages"]:
    # User's message
    st.markdown(f'<div class="chat-bubble-user">{message["user"]}</div>', unsafe_allow_html=True)
    # Bot's response
    st.markdown(f'<div class="chat-bubble">{message["bot"]}</div>', unsafe_allow_html=True)

Custom Styling

Custom CSS styles ensure the app is visually appealing and user-friendly. Example:

.chat-bubble {
    background-color: #007bff;
    color: white;
    padding: 10px;
    border-radius: 15px;
}

Secrets Management

The secrets.toml file stores sensitive credentials:

Langflow ID: Identifies the specific flow to connect.
Application Token: Authenticates API requests.
Endpoint: Specifies the Langflow backend entry point.

How It All Comes Together

The app performs these key steps:

Users submit a query via the Streamlit interface.
The query is sent to Langflow through the run_flow() function.
Langflow processes the query using your custom flow and returns a response.
The response is displayed in the Streamlit UI, styled as chat bubbles.

By understanding app.py, you can:

Customize its functionality for your specific use case.
Add more features, like new buttons or additional styling.

Next Steps: Experiment with app.py to expand your app's capabilities!

Name		Name	Last commit message	Last commit date
Latest commit History 28 Commits
.streamlit		.streamlit
assets		assets
.gitignore		.gitignore
Company_FAQ.pdf		Company_FAQ.pdf
Customer Support.json		Customer Support.json
README.md		README.md
app.py		app.py
requirements.txt		requirements.txt
sample_orders.csv		sample_orders.csv
sample_products.csv		sample_products.csv

melpadden/agentic-ai-workshop

Folders and files

Latest commit

History

Repository files navigation

Mastering Agentic AI with Langflow and Astra DB 🚀

Workshop Overview 🛠️✨

Key Technologies

What You’ll Learn:

Use Case: Empowering Customer Support with AI 🤝💡

Why This Use Case Matters

Workshop Goal 🎯

Prerequisites ✅

Let’s Make Magic Happen! 🦄✨

Step 1: Set Up Our Flow 🛠️🌟

Create a New Flow ➕

Choose a Blank Flow Template 📄

Name Your Flow ✏️

Step 2: Create a Simple Echo Flow 🔁💬

Add Components 🧩

Connect the Components 🔗

Test in the Playground 🎮

Step 3: Extend the Flow with an Agent 🤖💡

Remove Existing Connections ✂️

Add an Agent Component 🧩

Connect the Components 🔗

Configure API Credentials 🔑

Set Agent Instructions 📜

Test the Flow in the Playground 🎮

Step 4: Implement the RAG Flow 🔍📚

Add an Astra DB Component 🗄️

Connect the Components 🔗

Create a New Database in Astra DB 📋

Switch to Astra DB 🔄

Create an FAQ Collection 🗂️

Switch Back to Langflow ↩️

Configure the Astra DB Component ⚙️

Define Agent Instructions ✍️

Test the RAG Flow 🎮

Verify the Flow ✅

Step 5: Vectorize the FAQ 📚✨

Add Components to the Canvas 🧩

Connect the Components 🔗

Upload the FAQ Document 📥

Configure the Astra DB Component ⚙️

Run the Vectorization Flow ▶️

Test in the Playground 🎮

Verify Vectorization ✅

Step 6: Upload Sample Data 📦💾

Switch to Astra DB 🔄

Create the Orders Collection 📝

Create the Products Collection 🛒

Load Data into the Collections 📥

Verify Data Uploads ✅

Return to Langflow 🔙

Step 7: Build the Order Lookup Agent Flow 🛠️🔍

Add Astra DB Components 📦

Add an Agent Component 🤖

Connect Components 🔗

Validate the Flow ✅

Step 8: Combine Both Flows with a Manager 🛠️🤖

Rename the RAG Agent ✏️

Define ManagementAgent’s Role 📜

Connect Tools to ManagementAgent 🔗

Test in the Playground 🎮

Conduct Comprehensive Testing ✅

Step 9: Integrate with a Python Front End 🖥️🐍

Fork the Repository 🍴

Clone the Repository 📁

Set Up a Virtual Environment 🐍

Install Dependencies 📦

Configure Secrets 🔑

Retrieve Langflow Credentials 🔐

Run the Application ▶️

Test the Front End 💬

Understand How app.py Works 📝

Step 10: Deploy the Application to Streamlit Cloud 🌐☁️

Prepare Your Repository 📁

Log In to Streamlit Cloud 🌐

Create a New Deployment 🚀

Create the `Orders` Collection 📝

Create the `Products` Collection 🛒

Understand How `app.py` Works 📝

Appendix: Understanding `app.py` 📝

Packages