In this tutorial, you'll build a simple flow that reads data from a dataset, processes it with an AI prompt, and saves the results. Takes about 10 minutes.

Prerequisites

• An Evaligo account (sign up free at evaligo.com)
• Basic understanding of what flows are
• 5-10 minutes of time

Step 1: Create a Dataset

First, we need some data to process. Let's create a dataset with company names.

1. 1

   Navigate to Datasets Go to your project and click Datasets in the sidebar

2. 2

   Create New Dataset Click New Dataset and name it "Companies to Analyze"

3. 3

   Define Schema Add a variable called company_name

4. 4

   Add Sample Data Add 3-5 company names (e.g., "OpenAI", "Anthropic", "Cohere")

Step 2: Open Flow Playground

Navigate to Flows in the sidebar and click Flow Playground. You'll see a blank canvas with a node palette on the left.

Step 3: Add Nodes

3.1 Add Dataset Source Node

This node will read data from your dataset.

• Click Dataset (Read) in the left palette
• Select "Companies to Analyze" from the dropdown
• The node appears on the canvas

3.2 Add Website Mapper Node

This node will take company names and find their websites.

• Drag Website Mapper from the palette
• Place it to the right of your dataset node

3.3 Add Prompt Node

This node will analyze the website content with AI.

• Click Prompt in the palette
• You'll be prompted to select or create a prompt
• Create a new prompt with the template:

Analyze this website content and extract:
1. Main product/service
2. Target audience
3. Key differentiators

Website: {{url}}
Content: {{content}}

3.4 Add Dataset Sink Node

This node will save the results.

• Click Dataset (Write) in the palette
• Create a new dataset called "Analysis Results" or select an existing one

Step 4: Connect Nodes

Now connect the nodes to define the data flow.

4.1 Dataset Source → Website Mapper

• Click and drag from the Dataset Source's output handle (right side)
• Drop on the Website Mapper's input handle (left side)
• In the mapping modal that opens:
  • Map out.company_name → urlVar
• Click Save

4.2 Website Mapper → Prompt

• Connect Website Mapper output to Prompt input
• Map:
  • out.url → url
  • out.content → content

4.3 Prompt → Dataset Sink

• Connect Prompt output to Dataset Sink input
• Map out → in (the AI analysis will be stored as-is)

Step 5: Save Your Flow

Click Save As at the top and give your flow a name like "Company Website Analysis".

Step 6: Run Your Flow

1. 1

   Select Samples Click the Dataset Source node to open the dataset panel. Select 1-2 company samples to test with.

2. 2

   Click Run Flow Click the Run Flow button at the top. The flow will execute.

3. 3

   Monitor Progress Watch the progress bar and see each node light up as it processes. Node timing appears above each node.

4. 4

   Check Results When complete, click the Dataset Sink node to see the results in the dataset panel below.

Example flow: Dataset → Website Mapper → Prompt → Dataset

Understanding Execution

When you run the flow, here's what happens:

1. Dataset Source reads the selected samples (e.g., "OpenAI", "Anthropic")
2. Website Mapper finds their websites and scrapes content
3. Prompt sends the content to an AI model with your analysis prompt
4. Dataset Sink writes the AI's analysis to your results dataset

Next Steps

Improve Your Flow

• Add error handling - Nodes can fail gracefully; check failed status indicators
• Optimize the prompt - Use the Prompt Engineering Platform to test and evaluate your prompt
• Process more samples - Select all samples and run batch processing
• Add array processing - Use Iterator nodes to process lists within data

Deploy as API

Want to turn this into a production API? Add an API Input node at the start and deploy with one click.

Troubleshooting

Flow won't run

• Check that all nodes are connected (no floating nodes)
• Ensure all mappings are configured (click connections to edit)
• Select at least one sample from the dataset

Node shows error

• Click the node to see error details
• Check the mapping - ensure field names match
• Verify the prompt has all required variables

Results don't look right

• Test your prompt in the Prompt Playground first
• Check the Website Mapper output - does it have the right content?
• Try adjusting the prompt or using a different model