Pantry Rescue

Inspiration

Food waste often starts with good intentions. We buy fresh ingredients planning to use them, but busy schedules, forgotten groceries, and incomplete meal plans mean food can spoil before we get around to cooking it.

The challenge isn't always a lack of recipes. It's figuring out what to make with the ingredients we already have and buying only what's genuinely missing.

I built Pantry Rescue to connect those steps into one simple experience: rescue expiring food, plan meals around it, and get help shopping for the missing ingredients.

What it does

Pantry Rescue is a food-waste prevention and AI-assisted grocery shopping application.

The experience follows three steps:

1. Rescue — Start with what's going bad.

Users enter ingredients they already have, including quantities and expiration dates. Pantry Rescue identifies which foods should be prioritized.

2. Plan — Turn existing ingredients into meals.

The app recommends recipes that use ingredients approaching expiration. It prioritizes food rescue while considering recipe requirements and missing groceries.

Instead of creating an entirely new shopping list, Pantry Rescue subtracts available pantry quantities from recipe requirements, merges overlapping needs, and identifies only the remaining ingredients.

3. Shop — Approve the gaps and watch an agent help.

Users review and approve the missing ingredients, then choose a shopping destination.

Pantry Rescue integrates with Steel's cloud browser infrastructure to demonstrate an agent-assisted shopping workflow. In the verified Demo Market environment, the agent can search for products, present matches for approval, perform authorized cart actions, and check the resulting cart.

An experimental Walmart Canada integration also demonstrates live browser navigation and product searching, with human takeover when retailer verification or other obstacles arise. Real Walmart product matching and cart additions are not yet fully verified.

The user remains in control of product selection and checkout throughout the experience.

How I built it

I developed Pantry Rescue using React, TypeScript, Vite, and Node.js, supported by AI-assisted development workflows.

The application combines:

• Expiry-aware meal planning: Prioritizing recipes that help use ingredients before they spoil.
• Recipe discovery: Curated recipes alongside integration with TheMealDB.
• Pantry-aware ingredient calculations: Normalizing quantities, subtracting available ingredients, and identifying shopping-list gaps.
• Browser automation: Using Steel and Playwright to navigate shopping environments.
• Human-in-the-loop controls: Requiring explicit product approval before shopping actions.
• Cart verification: Checking observed cart contents rather than assuming an Add button click succeeded.
• Responsive interface design: A streamlined Rescue → Plan → Shop experience designed around a consumer's actual workflow.

I also used coding-agent skills and planning documents to guide development, implementation, and testing.

Challenges I ran into

Reliable grocery automation was the biggest technical challenge.

Retailer websites are dynamic. Product cards, pricing elements, search results, and page structures can vary, making product verification difficult.

During Walmart Canada testing, I encountered human-verification challenges, inconsistent search results, and cases where visible product listings could not be reliably converted into verified product matches.

Rather than treating uncertain results as successful actions, I designed the agent to stop safely, report what went wrong, and allow human takeover.

Another challenge was ingredient matching. Recipe quantities don't always use the same units as pantry entries, and similar grocery products aren't necessarily interchangeable. The application needed to avoid unnecessary purchases while keeping uncertain matches visible for review.

Accomplishments that I'm proud of

I'm proud of building a connected experience rather than three disconnected features.

Pantry Rescue begins with a real household problem and carries that information through recipe recommendations, ingredient calculations, grocery approvals, and assisted shopping.

I'm especially proud of the approach to human oversight and verification. The shopping agent is designed to show what it found, obtain approval, and verify outcomes rather than silently make purchasing decisions.

The project also pushed me to work through real browser automation challenges and design clear failure states instead of hiding them.

What I learned

This project reinforced that building with AI is about much more than generating code.

I learned how important it is to define a focused problem, break it into testable workflows, validate AI-assisted implementations, and distinguish what an agent attempted from what it actually accomplished.

I also learned that the best automation experiences don't necessarily remove humans from the process. They make routine tasks easier while preserving control over consequential decisions.

What's next

Future improvements could include:

• More reliable retailer integrations and live product matching.
• Optional receipt or fridge scanning to simplify pantry entry.
• Personalized recipe suggestions based on dietary preferences.
• Smarter ingredient substitution suggestions with user approval.
• Improved inventory tracking and reminders for food nearing expiration.

The long-term vision is to make Pantry Rescue a practical companion that helps households waste less food and shop more intentionally.

Built With

• css
• html
• node.js
• playwright
• react
• steel
• themealdb
• typescript
• vite