Meal Partner AI

Inspiration

Choosing what to eat after work is a small but repeated source of decision fatigue. When people are tired, they often do not want to search through recipes or compare a long list of recommendations. They simply need a few realistic options and a small push toward a decision.

I created Meal Partner AI to make that moment easier. The goal is not to create a perfect meal plan, but to help the user decide what to eat with as little effort as possible.

What it does

Meal Partner AI provides two practical meal options at a time through a short conversational interface.

It considers recent meal history, avoids suggesting the same meal repeatedly, and asks simple questions such as whether cooked rice is available at home. If the first suggestions do not feel right, the user can enter a short preference and receive two new options.

The interaction is intentionally short so that users can make a decision without reading long explanations.

How we built it

I built Meal Partner AI as a web application using React and TypeScript.

GPT-5.6 is used through the OpenAI API to generate two practical meal suggestions based on the user's current situation, short preferences, and recent meal history. Recent meal history is stored locally so the application can reduce repetitive recommendations.

Codex supported the development process by helping implement features, create and revise automated tests, investigate errors, and improve the conversational flow. I made the main product decisions, including limiting each response to two options, keeping questions short, and adding fallback behavior when an AI response cannot be used.

Challenges we ran into

The main challenge was balancing simplicity with useful personalization.

Too many questions would make the application feel like a detailed nutrition or recipe service, while too little context could produce repetitive or impractical suggestions. Another challenge was using recent history without making the conversation feel complicated.

Keeping the interface small while still providing meaningful suggestions required several revisions to the conversation flow.

Accomplishments that we're proud of

I completed a working MVP that combines AI-generated suggestions with recent meal history and a consistent two-choice interaction.

The application can guide a user from a vague request such as "What should I eat today?" toward a practical decision without presenting an overwhelming list of options.

I am also proud that this was my first AI application development project and that I was able to build, test, and connect the main features during the hackathon period.

What we learned

I learned that reducing choices can be more valuable than generating more information.

I also learned the importance of defining a narrow user problem before adding features. Automated testing and incremental development helped identify issues without changing several parts of the application at once.

Working with Codex made it possible to develop the application step by step while still keeping control of the product decisions.

I also learned that GPT-5.6 works best in this product when its role is clearly constrained: providing two usable options rather than an unlimited list of ideas.

What's next for Meal Partner AI

The next step is to improve personalization while keeping the conversation short.

Future versions could consider budget, available ingredients, supermarket purchases, dietary preferences, and basic nutritional balance. I would also like to improve the meal history system and make the application easier to use on mobile devices.

The long-term goal is to create a calm, practical decision-support companion for everyday meals.

Built With

• codex
• gpt-5.6
• node.js
• openai
• react
• typescript
• vite