Seattle residents have to sort their trash into compostable and non-compostable following guidelines specified by the city. It can be confusing, specially but not limited to someone moving here. This app simplifies that daily chore: snap a pic and let an app decide :-)
I turned the hard to resolve problem of recognizing diverse images of everyday items into a manageable solution combining Computer Vision, Natural Language Processing and Machine Learning techniques. The Hybrid recommender backend is developed in Python using Google Cloud Vision API, NLTK Lemmatizer, Scikit-learn Countvectorizer and Random-Forest classifier. I scraped 1,300+ pages from Seattle.gov to train Machine Learning models on telling apart compostable from non-compostables, then cleaned, lemmatized and vectorized the data. Then I tackled imbalanced classes via class weights, optimized the ML models and end-to-end validated the pipeline using real life user provided images. The HTML/CSS frontend uses Python Flask and Bootstrap. It has been deployed using the Google App Engine platform. The Notebooks here demonstrates key steps of developing this product and the project namesake directory contains the deployed codebase.
Please try it out !!!