SustainaBite - DAAARubi

Grocery Receipt scanning for seamless ingredient input
Actual Groceries/Ingrediant scan for seem
Inventory of expiring groceries/ingrediants
Recipe generation from soon to be expired ingredients
Carbon Coupon System

Inspiration

As we students are more accustomed to having meals prepared by someone else than doing so ourselves, we are not the best at keeping track of ingredients’ expiration dates. As a consequence, money is wasted and food waste is produced. This pattern not only affects us personally but also contributes to the global problem of food waste and its associated environmental impact. Additionally, as novices to cooking, we want a seamless guide on how we can fully cook the groceries that are going to expire

What it does

Our solution, "Sustainabite," is a user-friendly platform designed to seamlessly integrate into the routines of modern households for groceries. Here's how it works:

Groceries / Recipe Scan:

Users can either capture an image of their grocery receipt or directly photograph the items they've purchased. Our platform employs Document Understanding Transformers and Zero-Shot Foundational Vision Models to accurately identify limitless food items.

Smart Expiration Estimation:

Using database of perishable food as a reference material for an LLMs to generate an estimated expiration date for each food item

Personalized Recipe Generation:

To further encourage sustainable practices, Sustainabite automatically generates recipes based on the soon-to-expire ingredients. This not only helps users utilize their groceries efficiently but also enhances their culinary skills by suggesting creative recipes. Utilized edamam API, which is a repository of Food recipes to extract recipe and LLMs to condense information.

Accessible Grocery Management:

We provide users with a web-based application to manage their groceries. This application is seamlessly connected to a cloud-based database, ensuring users can easily track the status and estimated expiration dates of their items from any device.

Challenges we ran into

Trying to deploy large-scale foundational models (Donut, RecognizeAnything, GroundingDINO,CLIP) was a challenge due to expense costs. We managed to find a solution, running a 24GB NVIDIA RTX 3090 for less than 30 cents per hour.

Accomplishments that we're proud of

Utilization of Language Models

Integrating LLMs (Language Models) for both estimating expiration dates of ingredients and generating personalized recipes was a significant accomplishment. Leveraging the capabilities of these models allowed us to provide users with valuable insights into their groceries and inspire them to cook creatively with soon-to-expire items. This achievement demonstrated our aptitude for harnessing state-of-the-art AI technologies to solve real-world problems.

Micro-service Integration

Integrating multiple micro-services, including OCR, expiry date estimation, and recipe generation, into a coherent application was another challenge. Ensuring that these disparate components communicated effectively and produced accurate results required meticulous coordination. We learned the significance of designing clear API specifications protocols between services. This experience enhanced our understanding of scalable and modular software architecture.