By Vaibhav Verma
Foodie is a food review app that takes a dish-level approach to ratings. Friends always ask for the origin story of Foodie. Here is my attempt to tell it.
I first had the idea for Foodie during my first-year the UChicago. I was obsessed with the Chicago food scene and went downtown to try one restaurant after another. My freshman self would even write Yelp reviews, convinced that my palate was refined enough to warrant an opinion.
When I went to restaurants, I would always scour Yelp reviews, looking for what to eat. I would flip through menu page by page, trying to cross reference Yelp while groaning at just how many combinations of fried rice a Chinese place could have. I loved going to restaurants, but I was indecisive as hell and always wanted to order the best thing on the menu.
Because of this recurring experience, I was inspired to build an application that would have ratings for every dish on the menu. The goal would be to help users order the best dish, every single time.
Along with my friend Hojung, I started working on Foodie in my second-year. We gathered 1000 ratings via a Google Survey, built an iOS application and released it in the App Store.
It was a disaster. The app was clunky and the design was poorly thought out. The features included a Newsfeed, an Instagram-esque Profile Page, a search engine for both food and people, a list of restaurants ordered by distance, and the option for users to add restaurants themselves — we tried to do too much. Every part of the design from the Dish Rating Page to the Splash Page and the logo was yucky.
More crippling than the poor design or the cluttering feature set was the network effect required for Foodie to be useful in the first place. Our app did not have ratings to attract users, and it did not have users to drive ratings. It was a classic chicken-and-egg problem, and this problem seemed insurmountable.
We lost faith in the concept and soon stopped working on the project altogether. Hojung focused on new ventures, and I let the entrepreneurial bug fade, going through the motions of internships and completing my classes.
Two years later, I walked home after the first day of a machine learning class and thought, what if we could use ML to produce ratings for individual dishes algorithmically?
Essentially, we could use natural language processing and deep learning (sorry for the heavy use of lingo here) to build a sentiment analyzer. Then, we could take sentences from restaurant reviews (i.e. blog posts, social media) and pass them through the sentiment analyzer.
If a sentence contained a particular menu item, the sentiment analyzer could gauge the sentiment of that sentence and generate a rating for the particular menu item. In this way, we could generate dozens of ratings for every individual dish and skip the network effect problem entirely!
I called my friend Kathy, a phenomenal designer, and pitched her the project. Funny story — my roommate accidentally put plastic in the oven while cooking, and our apartment was filled in toxic fumes. Kathy and I didn’t care and talked for hours, creating the basic mockups of how the app would be laid out.
With Foodie in mind, I took a few more machine learning classes and started learning natural language processing independently. Tejas, an outstanding developer, also got onboard. Together, our team could take care of design, development, and the data.
Little by little, we made progress. It is now six months later. We have a growing team and just finished a three-week beta program with over 100 testers. We have over 14,000 ratings for ~150 restaurants and are adding ten new restaurants per day. Today, we even submitted the app to Apple, which means we are two weeks away from launching Foodie for the second time.
It will be exciting to see how users respond to this application. We have many more stories to share, and you will be able to find them here. If all goes well, this is where we will document, reflect, and describe our endeavor to make Foodie a reality.