Enhancing Workflows: No-Code and AI Integration Techniques
Written on
Chapter 1: Introduction to No-Code Automation
For more than a year, I've been developing various No-Code workflows that utilize AI technologies, such as ChatGPT, to accomplish several objectives:
- Automate tedious tasks
- Increase productivity
- Identify and filter essential information
One particular workflow I have frequently employed over the past year encompasses all these aspects, enabling automation of text extraction and categorization using LLMs like ChatGPT in a set-and-forget manner.
I am excited to share some Blueprints that will empower you to create these workflows yourself!
The No-Code platform we will utilize for these workflows is Make.com. While I will provide Blueprints specific to Make, similar outcomes can be achieved with tools such as n8n, Airtable, or Zapier. Today, we will focus on the fully automated extraction and categorization of information.
Section 1.1: Why Choose No-Code?
There are numerous advantages to opting for No-Code solutions. They allow for rapid development and iteration of automation tailored to the tools we already use. No-Code platforms can act as the connective tissue between various applications, functioning as a bridge without incurring significant effort or costs. Their user-friendly nature makes it feasible to automate tasks that previously seemed too complex.
While coding offers unparalleled flexibility, it also requires substantial expertise. Setting up integrations with services we use daily, such as Google Drive, Microsoft 365, or Notion, can be time-consuming, especially without knowledge of hosting, authentication, or different APIs. Many No-Code tools alleviate these challenges.
When combined with Generative AI like ChatGPT, the possibilities for automation expand significantly. As illustrated in an XKCD comic, it’s often more beneficial than you might expect!
Chapter 2: Automating Information Extraction and Categorization
The workflow we are implementing allows us to extract and categorize information from various sources such as emails, documents, and notes. This process can be incredibly useful for prioritizing leads, scoring product reviews, and organizing common complaints. With a straightforward prompt, we can efficiently process and categorize any text.
Instead of manually reviewing documents or messages, we can utilize LLMs to label, categorize, and analyze them effortlessly. For those eager to get started, the Blueprint available at the end of this article can be directly imported into Make.com!
Section 2.1: Workflow Overview
What exactly are we building? The goal is to create a workflow that automatically retrieves and structures information from text files placed in a designated Dropbox folder, subsequently storing the organized data in Google Sheets.
To achieve this, we will follow four key steps:
- Trigger: Define an event that initiates the workflow, such as uploading a document to Dropbox.
- Download the file: Access and process the file's contents.
- Extract information using AI: Employ a well-crafted prompt to parse the data.
- Store the information: Save the extracted data in a preferred format like Google Sheets or Excel.
In this example, I will utilize Dropbox for file storage, ChatGPT for parsing, and Google Sheets for data organization. However, feel free to adapt the workflow as needed.
Section 2.2: Setting Up the Trigger
Starting with a new scenario in Make.com, both Make and Zapier offer AI assistants that can significantly simplify the setup process.
The first step is to create the trigger. Select the folder you want to monitor and set the frequency for checks (the default is every 15 minutes). After uploading a file, you can run the Dropbox module to verify the upload. While this provides file information, we still need to access the content.
Section 2.3: Processing the File
Dropbox includes a "Download file" feature, which we will use to specify where the file should be downloaded. When we re-upload a file and execute the workflow again, we may encounter an issue where the data appears scrambled.
Fortunately, ChatGPT can guide us in resolving this encoding challenge. By searching for "Encoding" in the modules, we can locate a relevant tool to help us decode the data.
Section 2.4: Extracting Information with AI
To illustrate the extraction process, consider a fictional transcript between a sales representative and a prospective client. Extracting relevant information from such conversations can be labor-intensive.
By utilizing "JSON Mode" within Make, we can ensure a structured output. Both Mistral and ChatGPT support this format, making it easier to retrieve essential data.
For instance, if we prompt ChatGPT to identify participants, we may receive a generic response. However, with JSON Mode activated, the output becomes structured and manageable, allowing us to retrieve detailed information efficiently.
Now, we need to configure the OpenAI module. After entering your API key, select either the gpt-4-turbo or gpt-3.5-turbo model. Adjust the settings by setting the maximum tokens to around 1000 and a temperature of 0.2. Lastly, ensure that the response format is set to JSON object.
A prompt template for extracting information can be constructed as follows:
Instruction:
Your task is to extract information from the following transcript. Be concise, ensuring separation between "Sales Rep" and "Potential Client." Adhere strictly to the provided format.
Example:
{
"sales_rep": {
"name": "Alex",
"company": "DreamTech Solutions"
},
...
}
Transcript:
{{5.data}}
Section 2.5: Storing Information in Google Sheets
The final step is straightforward. Now that we have a JSON output, we need to parse it effectively. Using a JSON-specific module, we can process the output from the ChatGPT module.
Upon running the workflow, the output should be formatted correctly and ready for storage in Google Sheets.
Now, simply link to the appropriate file in Google Sheets. If everything has been set up correctly, you can easily map the values to the respective columns.
With this completed, you can explore further enhancements. By adding routers and filters, you can create diverse functionalities, such as task management or automated emails based on specific conditions. The potential applications are vast, allowing for continuous improvement of your workflows.
If you'd like to explore what I’ve created, the Blueprint is available for download. Import it into Make.com, log in with your credentials, and set up your OpenAI account to begin automating your repetitive tasks!
Blueprint can be downloaded here: Download link