Skip to main content

AI-Powered Smart Cooking Assistant Using Generative AI for Personalized Recipe Recommendation and Nu

Page 1

International Research Journal of Engineering and Technology (IRJET)

e-ISSN: 2395-0056

Volume: 13 Issue: 06 | Jun 2026

p-ISSN: 2395-0072

www.irjet.net

AI-Powered Smart Cooking Assistant Using Generative AI for Personalized Recipe Recommendation and Nutrition Analysis Pallavi Kumari¹, Vivek Yadav², Riya Priyal³, Priyashi Priya´ ¹²³´Department of Computer Science & Engineering, Shivalik College of Engineering, Dehradun, India ---------------------------------------------------------------------***---------------------------------------------------------------------

Abstract - Every day millions of people open their fridge,

calories it has? Why not let someone plan meals for the whole week instead of just one meal? Why force someone to stare at a screen while their hands are covered in flour when the phone could just read the steps aloud?

stare at whatever is inside, and have no idea what to cook. It seems like a small problem but honestly it leads to a lot of food being wasted, a lot of unhealthy takeout orders, and people just making the same three dishes over and over. We built this project because we felt like AI could actually help with this in a practical way, not just as a concept but as something you could actually open on your phone and use. The system we built lets a person type in whatever ingredients they have, pick a cuisine style, choose their diet type, and get back a complete recipe that the AI generates on the spot. Not something pulled from a database actually generated fresh based on what was asked. Along with the recipe, the app shows a nutrition breakdown so the user knows what they are eating. We also added a 7-day meal planner because one recipe is not really enough if someone wants to eat better across a whole week. And there is a voice feature so that while you are cooking you can just listen to the steps instead of looking at the screen with messy hands. On the technical side we used React and Next.js for the frontend, Node.js for the backend, MongoDB to store data, and the Gemini and Claude AI APIs for the recipe generation part. This paper explains how we built all of this, shows screenshots of the actual working app, and talks honestly about what worked, what has problems, and what we would improve given more time.

So the project grew from a basic recipe generator into something more complete. It is a web application with four main things it can do: generate AI-powered recipes from ingredients, show nutrition information for every recipe, plan a full 7-day meal schedule based on health goals, and support voice input and output for hands-free cooking. We are not claiming this is a perfect product. But it works, it is genuinely useful, and it shows what can be done when you connect generative AI to a real everyday problem. The rest of this paper covers related work in Section 2, explains the system design in Section 3, goes through each module in Section 4, shows the actual results with screenshots in Section 5, talks about limitations in Section 6, and concludes in Section 7.

2. LITERATURE REVIEW

Key words: Generative AI, Recipe Recommendation, Nutrition Analysis, Meal Planning, Large Language Models, Voice Assistant, MongoDB, Gemini API

1. INTRODUCTION When we first thought about what project to build for our final year, we kept coming back to everyday problems that technology has not really solved well yet. Cooking felt like one of those. There are thousands of recipe websites, apps, food blogs but none of them actually answer the question a person has in real life, which is: I have these specific things in my kitchen right now, what should I make? Most apps just show you recipes from a database and you have to filter and scroll and hope something matches. That felt like a missed opportunity.

Some researchers tried collaborative filtering instead, the same approach Netflix uses for movies. The idea is that if many people who liked recipe A also liked recipe B, then recommend recipe B to others who liked recipe A. All Recipes uses something like this. The problem for cooking specifically is the cold start issue a new user has no history, so the system cannot personalize anything. Also the recommendations are based on taste preferences, not on what ingredients you actually have available, which is often the more pressing constraint.

Generative AI changes what is possible here. Instead of searching through existing recipes, the system can actually create one based on whatever the user has. We thought that was interesting enough to build a full project around. And once we started building, other features made sense to add. Why show a recipe without telling someone how many

© 2026, IRJET

|

Impact Factor value: 8.315

Recipe recommendation is not a new research topic. People have been working on it for at least fifteen years. The early approaches were mostly content-based filtering you tell the system your dietary restrictions, it searches a big database and shows you matching recipes. Platforms like Yummy became popular this way. It works okay but the fundamental problem is that it can only show you what already exists in its database. There is no creativity, no adaptation to unusual ingredient combinations, and the personalization is quite shallow.

Things changed quite a bit when transformer-based language models came along. Bien et al. published RecipeNLG in 2020, which was a large dataset of recipes used to train models that could generate new recipes from

|

ISO 9001:2008 Certified Journal

|

Page 347


Turn static files into dynamic content formats.

Create a flipbook
AI-Powered Smart Cooking Assistant Using Generative AI for Personalized Recipe Recommendation and Nu by IRJET Journal - Issuu