Our inspiration came from a worldwide problem. People waste food because they go grocery shopping when they can make a perfect meal from what they have at home and a lot of food is potentially wasted. We decided to tackle this problem by making it easy for people to discover what they can cook from the food they already have. With Recipe Wizard, you can discover new recipes purely based on grocery receipts and food in your pantry.
What it does
Recipe Wizard enables you to scan receipts of stuff you've just bought and asks questions about what you have in your pantry. It then recommends you recipes you can make entirely with food you already have!
How we built it
We were working on two parts concurrently, the front end written mostly using JQuery and the python backend using Flask. We used the Google vision API to extract information from receipts and spacy for NLP. We wrote our own algorithm for selecting which ingredients to ask about next.
Challenges we ran into
- The mapping of natural language on grocery receipts to ingredients.
- Scanning receipts and extracting only correct, important information from them.
Accomplishments that we're proud of
During this hackathon, we managed to gather data for 20,000 recipes so the system is easily growable and expandable.
We construct several large tries which we use to efficiently search for the most relevant recipes given your pantry and grocery receipts.
We are happy that we built a pipeline which scans receipts, extracts food items, and adds them to your pantry. The questions about ingredients based on our own algorithm, asking about the the ingredient which will enable you to make the most recipes with what you already have.
What we learned
JQuery is awesome as we managed to implement a lot of functionality on the front end in relatively little code.
Advanced web scraping for recipes
Working with and acquiring a large dataset (20,000 recipes)
NLP training and use