Unlock the Secrets of Books with Tana and OpenAI ChatGPT

Table of Contents

1. Introduction
2. Building a Complex Workflow
3. Enabling Full Text Extraction
4. Setting Up the Fetch Full Text Command
5. Parsing HTML to Extract Relevant Information
6. Asking Questions about the Book
7. Generating AI-Powered Responses
8. Structuring the Answer
9. Tuning the Context Window Percentage
10. Publishing the Workflow as a Template

Building a More Complex Workflow with API and Batch Prompt Processing

In this article, I will guide You through the process of building a more complex workflow using API commands and batch prompt processing. We will be using the API command to extract the text of an article or book, and a feature called batch prompt processing to extract relevant information. By the end of this article, I will provide you with the workflow template that you can easily install and use.

Introduction

When working with API commands, it can be challenging to extract relevant information from a large amount of text. Batch prompt processing allows us to break down the text into smaller chunks and send them to the AI for analysis. This helps us extract valuable insights and generate structured summaries. In this article, we will go step by step in building a workflow that utilizes these features.

Building a Complex Workflow

To start off, we need to Create a new tag and add the necessary fields. The tag will be called "Ask Questions of Book," and we will add two fields: URL and Full Text. The URL field will be used to fetch the full text of the book or article, while the Full Text field will store the extracted text for further processing.

Enabling Full Text Extraction

The first step is to enable the fetching of full text from a given URL. To do this, we need to add a button to the tag and configure it to make an API request. The URL field will be used as the value for the URL parameter in the API request. Once the API retrieves the HTML from the URL, we will use the "Parse Results: HTML to Tunnel Paste" command to extract the relevant text and convert it into Tunnel Paste format. Finally, we will store the output in the Full Text field.

Setting Up the Fetch Full Text Command

To fetch the full text, we will use the "Fetch Full Text" command. We can add a node filter to ensure that the command only appears when the Full Text field is empty. This way, we avoid unnecessary API calls if we already have the full text. By configuring the command with the URL field as the URL parameter, we can fetch the full text and populate the Full Text field.

Parsing HTML to Extract Relevant Information

After fetching the full text, we need to extract relevant information from it. We will use the "Ask AI" command to ask a specific question and get relevant quotes and excerpts. The "Batch Prompt Context" feature will allow us to split the text into smaller chunks that fit within the context window. By referencing the Full Text field in the batch prompt context, we can refer to it using the sys context variable. This allows us to write a prompt that extracts information from the full text.

Asking Questions about the Book

Next, we need to ask a question about the book or article. We will add a question field and use it as the prompt for the "Ask AI" command. By referencing the Full Text field in the prompt, we can ensure that the AI extracts information relevant to answering our question. The AI will provide us with extracts and direct quotes from the text that address the question.

Generating AI-Powered Responses

Once we have the extracts and direct quotes, we can generate an AI-powered response. We will create a new field called "Answer" and add a button to the tag to trigger the generation of the response. By configuring the button's command with the prompt and target node, we can generate an answer Based on the extracted information. The response will be stored in the Answer field.

Structuring the Answer

To structure the answer, we can use hierarchical markdown. We will provide a template that guides the AI in organizing the extracted information into a comprehensive and well-structured answer. By including all the relevant information provided by the AI, we can create a concise and informative summary of the book or article. We will also include instructions to ignore any sentences about information not available to avoid confusion.

Tuning the Context Window Percentage

To optimize the workflow, we can tune the context window percentage. By adjusting the fill context window percentage, we can ensure that the AI has enough context to generate Meaningful responses. In our case, we can increase the percentage since We Are sending a large amount of text and only expecting a few sentences in response. This reduces the number of API requests and improves the performance.

Publishing the Workflow as a Template

Once we have built the workflow, tested it, and made any necessary improvements, we can publish it as a template. This allows others to easily install and use the template in their own workflows. By sharing the template link, we can promote collaboration and encourage others to build upon our work.

In conclusion, building a complex workflow with API and batch prompt processing enables us to efficiently extract relevant information from large amounts of text. By utilizing features such as fetching full text, parsing HTML, asking questions to AI, and generating structured responses, we can create powerful workflows that provide valuable insights. The published template ensures that others can benefit from our work and contribute to further improvements.

Highlights

• Build a complex workflow utilizing API commands
• Fetch full text from a URL and convert it to Tunnel Paste format
• Extract relevant information using batch prompt processing
• Ask questions and generate AI-powered responses
• Structure the answer using hierarchical markdown
• Optimize the workflow by tuning the context window percentage
• Publish the workflow as a template for others to use

FAQ

Q: Can I use this workflow for articles as well? A: Yes, you can use this workflow for both books and articles by providing the URL or full text accordingly.

Q: How accurate are the generated responses? A: The accuracy of the generated responses can vary depending on the quality of the input text and the prompt used. It is recommended to experiment and fine-tune the prompt to get the desired results.

Q: Can I modify the workflow to suit my specific needs? A: Absolutely! The workflow is designed to be flexible and customizable. Feel free to modify it according to your requirements and use case.