TasteGPT: The key to Tastewise’s best insights

May 3, 20245 min
Axel Strubel photo
Axel Strubel

TasteGPT, in many ways, holds the key to unlocking the Tastewise platform’s packed abundant information designed to help users understand food trends, diets, motives, and more. However, many of our most advanced features are often underutilized or difficult for newcomers to discover.

The platform is so comprehensive and contains such extensive information that we needed to devise a method for guiding users toward actionable insights tailored to their needs!

The platform features various tabs and pages, including social media posts from Instagram for inspiration. This article will explore how we integrated ChatGPT into Tastewise’s platform to improve user experience and simplify navigation.

The Challenge

With numerous tabs and pages related to diverse food trends and information, it can be challenging for users to quickly find the answers they seek without extensive knowledge of the platform’s layout. Our goal: To create a more intuitive user interface that understands and responds to natural language queries, making it easier for users to find what they’re looking for.

Integrating ChatGPT

To achieve this, we used ChatGPT, an AI model trained on conversational data, which excels at understanding and responding to human instructions. We created prompts that generate text for navigating within the platform, allowing users to ask questions and receive URL parameters and page information to be redirected accordingly.

One Prompt to Rule Them All

We started by giving ChatGPT a single prompt that explained the different components of Tastewise’s platform, including various tabs, pages, and key terms. We then set up the model to interpret and categorize user requests and respond with a structured data format. An always good starting point for understanding pages in Tastewise is the glossary.

Since the results were good but could be improved we tried to give many questions and answers to ChatGPT to fine-tune the model. It turned out that after a few versions of the training we didn’t see great results by doing so in comparison to writing a big prompt.

Building the Prompts by simplified short examples:

In the end, we decided that the prompt approach was interesting and we gave ChatGPT a series of prompts that explained the different actions we need from it. The advantage would be that each prompt would specialize in one particular aspect. To get into more detail let’s look into parts of these prompts :

Explaining Technical I/O to chatGPT

Respond to requests sent to a chat integrated into a food analytics platform which will be interpreted by a server to know which is the most relevant page to show

Explaining different pages to chatGPT

Some examples of what could be included in the prompt :

Explanation of the pages:

 “Spotlight”: a page used to validate a concept, to know anything about it including the life cycle stage (is oat milk trending?), also in case of wanting  an overview or a report

This part of the prompt explains what our spotlight page is for and what it could be used for in a few lines.

  1. Creating filters so that the user query will also incorporate important filtering capacities 

“audiences”:  Should be extracted from the user query. Only the following ones are accepted : “baby boomers”, “beers lovers”, “chefs”, “coffee lovers”,  “fathers” , “female” ,

“foodies”, “gen x”, “gen z”, “health”, “influencers”, “keto”, “male”, “millennials” , “moms”,  “top chefs”, “vegan”, “vegetarian”.

Here all our main audience filters are described to the prompt of chatGPT

The result, then, of the question “Is oat milk trending among Gen Z?” will be :

Handling User Queries

Once the AI model had been trained, we integrated it with Tastewise’s platform to handle user queries. Users can now ask questions in natural language, and we receive responses from ChatGPT containing the URL parameters and page information needed to redirect them to the appropriate section of the platform.

Example Query:

User: What are popular flavors for Non-Alcoholic Drinks among Gen Z?

Example Response from ChatGPT:

Users should be redirected to the Ingredient Page with search terms that will be non alcoholic drinks and filters will be audience “gen_z”. 

Which would be then interpret like this :

Implementing the Solution

We integrated ChatGPT responses into Tastewise’s platform, enabling seamless navigation and improved user experience via TasteGPT. Users can now ask questions in natural language and be redirected to the relevant sections of the platform, making it easier to discover insights and inspiration. We are also adding internal tools to ensure that the query will be translated accurately into the platform’s appropriate categories.

By integrating ChatGPT into Tastewise’s food analytics platform, we have created a more intuitive and efficient user experience. Users can now use TasteGPT to navigate the platform using natural language queries, simplifying finding relevant information and insights. This innovative approach to enhancing user experience can be applied to other platforms and industries, showcasing the potential of AI in improving user interfaces and transforming the way we interact with technology.

What can food intelligence do for you?