Lessons Learned From Integrating ChatGPT into a Low Code/No Code App

The time for you to design custom AI applications that simplify your workload and boost your productivity is here. No longer do you need a master's degree or expensive algorithm to get started. A question, some examples, and an understanding of how the settings work with a specific GPT are more than enough to build a custom agent capable of recommending fine dining on your next vacation, or even providing tips, explanations, and best practices for your actual code. Here I list a few lessons I gained while creating a custom ChatGPT integration for a client. The purpose of this integration was to summarize pertinent information for a specific audience and then produce a customized template based on that knowledge.

1. Gather your tools. 

   1. I was already working with a client on a Bubble app for this integration. Bubble is known for its ease of use, as well as versatility as a no-code tool. In Bubble, templates are available for customization for a price, but for this integration, all needed is a text input area, a few buttons for different API calls, and a text display area for the output. 

2. Gather your credentials

   1.      ChatGPT/Open API credentials 

      1. Without an account and buying tokens, you aren’t able to run ChatGPT transactions.

   2. I also used the Bubble plug-in,  API connector. This is for the external API calls to ChatGPT, and also for gathering the information I intended to pass on to the GPT to answer my questions.

   3. I also using a web scraping service for this particular implementation. 

3. Know your limits

   1. Limits on the data: 

      1. There is a max on the number of characters that can be sent in a request such as if you’re providing an example/input. 

      2. Try to find the limit as early as possible so you can also look for ways to navigate around it and still have the API meet your needs

      3. If there is a limit, get the most relevant, necessary content first. This will shape your responses tremendously.

   2. Limits on the tokens:

      1. Also impacts the usefulness and verbosity of the response. 

      2. More tokens, more verbose response. 

4. Initiate your entrance point into your integration

   1. Set up a workflow:

      1. Ask a question

      2. Enter a prompt

      3. Data inputs needed 

      4. Call to API 

      5. Connect this to your buttons, text input area, and text output area through workflows in Bubble 

5. Adjusting your responses to make them more valid

   1. Adjusting responses for consistency: 

      1. If I like a part I response, I would record it and all input parameters, and temperature. These become your presets or default prompts. 

   2. Adjusting for more creative responses 

      1. Using settings like temperature where a higher temperature indicates a more creative response, while a lower temperature is better for logical, fact-based, and consistent responses.

   
   
   
   

   Additional Resources:

   • Connect ChatGPT with Bubble

   • Connect Bubble with ChatGPT Youtube

   • Chat GPT Settings