Delivering Novel Software Experiences utilizing Machine Studying & AI


The vacation season is upon us and with it, all of the wealthy and scrumptious meals that heat our our bodies and problem our waistlines. At Databricks, we wished to have somewhat enjoyable and share with our clients some wonderful recipes to make the vacations much more pleasant. However this being Databricks, we needed to go and accomplish that by leveraging AI.

To indicate the simplicity of AI with Databricks in motion, we determined to resolve a top-of-mind problem: How can we make the very best pumpkin pie? This weblog submit walks via how we used the Databricks Serverless Actual-Time Interface with highly effective open supply applied sciences to construct an AI-driven search app.

Meet the Databricks Vacation Recipe Finder app

The Databricks Vacation Recipe Finder app is a straightforward, AI-driven utility that searches for meals gadgets primarily based on a picture or descriptive textual content supplied by a consumer. With this app, you’ll be able to search the Epicurious Meals Elements and Recipes knowledge set utilizing both image-based search. Add an image of your favourite vacation meals merchandise – have we talked about pumpkin pie? – or click on on any of the randomly chosen photos to search out recipes for meals related in look to your chosen photos.  Alternatively, you need to use the search field to search out recipes primarily based on descriptive phrases and phrases.

The inspiration for the applying was the widespread problem of seeing a dish however not fairly understanding what it’s.  We envisioned – no pun meant – a easy search the place you would add a picture of that arduous to establish dish, and Databricks might search an enormous catalog of meals gadgets to search out one thing related.  For every prompt merchandise, you’ll be able to see a picture that will help you get a visible sense of the dish, in addition to the components and directions for its preparation. We’ve already recognized various dishes we’ll be including to our vacation tables this 12 months, and we invite you to do the identical.

How we did it

For the app, we leverage over 13K recipes and related photos scraped from the Epicurious web site and made accessible as a part of this Kaggle dataset. The image-based search we’ve employed is predicated on the idea of embeddings. With out getting too technical, you’ll be able to consider an embedding as a brief, mathematical illustration of a picture or a block of textual content. After we evaluate the embeddings related to two gadgets, the space between their two embeddings tells us one thing in regards to the similarity between them.

If that sounds sophisticated, it needn’t. All of the logic and mannequin coaching required to calculate such embeddings has been tackled by the open supply neighborhood. Utilizing Databricks, all we have to do is obtain an appropriately pre-trained mannequin with a single line of code and use it to arrange the information we want to search.

As soon as the information is ready, sometimes, the problem could be integrating the mannequin into our utility. However as soon as once more, Databricks has us coated. Leveraging Databricks’s mannequin serving capabilities, we will wrap our mannequin with a easy API and deploy it to a scalable REST endpoint. It’s a couple of traces of code adopted by a pair clicks within the Databricks UI, and we’re up and operating.

Give the Vacation Recipe app a strive 

It should take us extra time to bake a pumpkin pie than it did to write down this utility, and that’s actually the entire level. At Databricks, our objective is and at all times has been to simplify using machine studying and synthetic intelligence. We’ve purposefully constructed our platform with options meant to make not simply the processing of knowledge and coaching of fashions simpler, however to make sure the insights derived via these could be built-in again into enterprise functions. 

We invite everybody to work together with the applying, which we are going to host all through the 2022 vacation season, and to obtain all of the code utilized in its preparation from our public GitHub repository. As you take pleasure in scrumptious dishes this vacation season – a few of that are hopefully impressed by the app – we invite you to consider all of the methods you would possibly leverage knowledge and AI in your functions this coming 12 months. 

To all our clients, we want you content holidays! And for individuals who use the app, we’d like to see what you’ll be cooking! We invite you to share your findings with the hashtag #DatabricksRecipes! 


Please enter your comment!
Please enter your name here