Document-Based AI

What is this feature?

An AI pipeline that analyzes large text data and quickly retrieves essential information.

Use it for knowledge management, summarization, and decision support.

Use Case Example

Structure of Document Retrieval Q&A

First, sign up or log-in to WindyFlo.

1. Create a new pipeline.

1) On the menu, go to My Pipelines and click 'Create Pipeline.'

2) Fill in the pipeline details, and

click Create AI Model Pipeline.

• Pipeline Name: Name your pipeline.
• Description: Summarize the purpose of the pipeline.
• Tags: Add tags to make your pipeline easier to discover.
• Public/Private: Choose whether to keep the pipeline public or private.

2. Build your pipeline.

1) Select Pipeline on the left-side menu.

This will open your workspace for this pipeline.

2) Click Add Node, search the node you need, and drag it onto the workspace.

Node List :

• Text File or Pdf File
• Recursive Character Text Splitter
• In-Memory Vector Store
• OpenAI Embeddings
• Conversational Retrieval QA Chain
• ChatOpenAI
• Buffer Memory

3) Connect the nodes.

Node connection order :

1. Recursive Character Text Splitter → Text File & Pdf File
2. Text File & Pdf File → In-Memory Vector Store
3. OpenAI Embeddings → In-Memory Vector Store
4. In-Memory Vector Store → Conversational Retrieval QA Chain
5. ChatOpenAI → Conversational Retrieval QA Chain
6. Buffer Memory → Conversational Retrieval QA Chain

4) Configure the node parameters.

1. Recursive Character Text Splitter
   • Chunk Size : 1000
   • Chunk Overlap : 200
2. Text File & PDF File
   • Upload File : Upload a TXT or PDF file.
3. OpenAI Embeddings
   • Connect Credential : Enter an external service API.
   • Model Name : text-embedding-3-small
4. ChatOpenAI
   • Connect Credential : Enter the external service API.
   • Model name : Select GPT-4 (latest) for smooth conversation.
   • Temperature : Set to 0.7 for creativity and consistency in responses.
5. Buffer Memory
   • Memory Key : Chat_history(default)
   • Input key : input(default)

3. Save and test-run.

1) Click Save, and then click Run.

2) Type a command to test the pipeline.

Example:

{Please provide the business information.}

4. Share the Pipeline

Share via API

1) Save your pipeline, and then click 'Embed as API.'

To make it accessible via API or embedded on a website.

2) Choose your preferred language (HTML, React, Python, JavaScript, or CURL) and copy-paste the generated code into your service.

Use as chatbot

1) Alternatively, to share it as a chatbot with no additional integration, select Share Chatbot.

• This feature is only available for pipelines that function as chatbots.
• Pipelines designed for data processing may not support this feature.

2) Customize the chatbot settings (e.g., title, welcome message) and click the new tab icon to launch.

• Left icon: Copy link
• Right icon: Open link

3) Test your chatbot by entering a query in the "Type your question..." field.

• Once testing is complete, copy the link and share it with your audience!