Hello AI Enthusiast,

If you've ever tried to eat healthier without overhauling everything at once, you know how hard it is to find practical, food-specific guidance. Jonas Wenke, a participant in our AI Agent Bootcamp, built a tool to solve exactly that.

Many chronic diseases are linked, at least partly, to insulin resistance caused by persistently high blood glucose levels. One way to address this is eating food with a lower Glycemic Index (GI), a measure of how quickly a food raises your blood sugar. The problem is that transitioning to a low-GI diet isn't straightforward. It requires knowing the GI of hundreds of ingredients, finding suitable alternatives that still taste good, and doing this consistently meal by meal.

Jonas wanted to lower his blood glucose levels gradually, without flipping his diet upside down overnight. His manual process involved researching GI values, hunting for low-GI recipes, and swapping ingredients on the fly, but it was unstructured and time-consuming.

So he built CandyDetox GI Swap: a WhatsApp assistant that takes any meal idea or ingredient list and returns the most impactful low-GI swaps, ranked and explained.

The core of CandyDetox is a well-crafted system prompt that turns an AI model into a personal nutrition advisor focused entirely on glycemic index. Jonas connected it to WhatsApp using Make, so the whole thing works inside an app he already uses every day.

Here's the flow:

🥣 User sends a meal idea or ingredient list via WhatsApp

🤖 AI assesses the GI of each ingredient and identifies the highest-impact swaps

📋 WhatsApp sends back a ranked list of alternatives, sorted from most to least impactful

In Make, create a new scenario with three modules:

The system prompt defines how the AI behaves every time someone sends a message. It tells it what role to play, what to do with the input, and exactly how to format the response.

Here's a template you can adapt for your own use case:

Once the automation is live, test it with a few real meal ideas. Jonas found that tweaking the formatting instructions, specifically asking for bold headers and no concluding sentences, made the output much more scannable. He also tested two versions of the prompt and compared quality scores and token costs, landing on a version that was both more accurate and cheaper to run.

Prompt formatting matters as much as content. Small changes, like specifying "no full sentences" or "sort from lowest to highest GI," had a big impact on output usability.

Test with real inputs. Jonas scored both prompt versions against actual meal descriptions. The second version scored higher on quality and cost less per completion.

Scope down first, then expand. Jonas started with the core swap logic working before wiring everything together. Getting the prompt right before connecting the automation saved a lot of back-and-forth.

Pick one recurring meal you eat during the week. Describe it in plain language to ChatGPT and ask it to identify the three highest-GI ingredients and suggest lower-GI alternatives that keep the dish recognizable. Note how much the suggestions would actually change what's on your plate.

Disclaimer: AI-generated nutritional suggestions are a starting point, not medical advice. Always consult a healthcare professional before making significant dietary changes, especially if you have a medical condition.

Want to get even more practical? Explore hands-on AI learning with AI Academy:

We'll be back with more AI tips soon!