Introduction

HowHot is a web application designed to help users assess the spiciness of meals from just an image. While some people love spicy food, others—like myself—may need to avoid it. I realized that there is no common measurement for spiciness that everyone knows. The Scoville Heat Unit (SHU) is one such metric, but it’s primarily used for raw peppers and isn’t commonly used when talking about food with friends.

In addition, it’s often difficult to determine how spicy a meal is just by looking at it—or even by reading its name on a menu.

This project aims to solve that problem by allowing users to upload a picture of their food and receive a predicted spice level on a 0 to 5 scale, similar to restaurant review apps. The application uses a deep learning model trained on real-world food images to estimate spiciness. After seeing the model’s prediction, users can also provide feedback to help improve future predictions.

With this tool, users can make safer and more informed food choices.

Here’s the link to the app: https://howhot.netlify.app/

https://howhot.netlify.app/

System Architecture Diagram

The HowHot application is built using the following architecture:

• Frontend: Built in React and deployed using Netlify for fast, global delivery.
• S3: Stores uploaded food images temporarily for processing.
• Prediction Service: A backend hosted on Railway runs the uploaded image through a custom-trained model and returns a spiciness score.
• Feedback Mechanism: Users can respond to the predicted level using a “Not Quite?” button, selecting the correct spice level.
• PostgreSQL: Logs feedback data including predicted and corrected spice levels for future retraining.
• Continual Learning: model adapts based on user feedback to make increasingly accurate predictions over time.

AI Model and MLOps