Three problems we are trying to solve at Zomato
Our engineering team is always cranking to improve our services and make your food hunt easier. As a small update, here is what's keeping us awake at night these days:
Building a dish database
Wouldn't it be great if we could search for restaurants serving a specific dish? For example, it would be great to be able to find the best places that serve a Tenderloin Steak. The only way to make this happen is if we have structured data for every dish served in a restaurant. We've already made some effort in this direction the manual way - by typing out a few thousand menu cards in every city that we are in. However, given the churn rate of the restaurant industry (about 25% restaurants shut down every quarter), it is getting difficult for us to manually keep pace with all the new menu cards that come our way. We are trying to create intelligent OCR systems to supplement our existing dish ontology. Misspellings (in the menus) and from the OCR mechanism just make it all the more complicated.
Spam control
Two years ago, this wasn't a problem. Why? Very simply, a restaurant's rating Zomato didn't make or break its business back then. Nowadays, it does make a difference, to a certain extent. Since a lot of restaurant owners also realise this, they try to game the system by planting fake reviews on the platform. We have, over time, learnt how to identify most of these cases, but some intelligent spam tactics skip our automated filters. Our filters currently range from simple data analysis and pattern matching to user behaviour mapping over a few months. However, we realise that we constantly have to keep track of what is making its way onto the platform to help maintain its neutrality and usefulness to our users. Also, we need to evolve these systems to outsmart everyone who tries to game the system (despite following the rules). It's an ongoing effort, and I don't think we will ever be able to say this problem has been solved for good.
Lower server latency for our geographically widespread traffic
This isn't something new, and a few companies have already done this very efficiently. Most companies our size don't even care about this. However, we want to serve all our traffic (in the Philippines, as well as in London) from servers that sit close to our users' physical locations. Lower server latency = users get to food faster. Replicating all our infrastructure in various locations on a master-slave basis and then keeping everything in sync is not easy for us as of now, and this is something we are trying to address over the very short term itself. Do any of these challenges excite you? Do you think you can help us solve these issues? If yes, we are looking for you. Email me at [email protected]
