DEV Community

Rob Lauer for Progress

Posted on

Using ML Kit for Firebase to Tap into Beer Ratings

Over the past year machine learning (ML) has gone mainstream. How do we know this? When we let computers create paint colors:

generated paint colors

Much like with augmented reality, we are still collectively working out the most practical applications for ML. But the best use of any new technology solves an existing real world problem. For me, the king of (first world) problems is deciding which beer to drink.

How many times have you entered a restaurant or bar and been assaulted with a beer menu of copious choices, about 5% of which you've heard of?

confusing beer menu

I decided it was in my palette's best interest to create a mobile app that would allow me to leverage:

Why NativeScript?

NativeScript is a free and open source framework that allows you to create truly native mobile apps for iOS and Android from one codebase. You use the web skills you already possess (JavaScript/TypeScript/Angular/Vue.js, CSS, and an HTML-like markup language) to create performant apps.

If you've heard of React Native, you can think of NativeScript as React Native, but for Angular or Vue.js

I'm a big fan of skill re-use (ain't nobody got time for Swift AND Java development these days!) and I'm also a JavaScript purist, so NativeScript is a natural fit for me.

As a quick example, you can see how NativeScript's markup syntax lets you leverage native UI elements via one consistent markup language:

    <Page loaded="pageLoaded">
        <Label text="Hello ML Kit!" class="content" />
        <Button text="OK!" />
Enter fullscreen mode Exit fullscreen mode

...and the same goes for CSS:

    Button {
        font-size: 32;
        color: orange;

    .content {
        font-family: Lato;
        background-color: purple;
Enter fullscreen mode Exit fullscreen mode

Why ML Kit?

Even though machine learning is still in its infancy, we've learned to trust Google in its infinite wisdom with bleeding-edge technologies. It doesn't hurt that NativeScript has a robust Firebase plugin, including support for ML Kit!

nativescript firebase logos

If you're not familiar with ML Kit, it's an SDK that unleashes the power of Google's machine learning capabilities to mobile apps. It's relatively easy to integrate and requires no knowledge of neural networks to get rolling.

As a mobile-focused machine learning SDK, ML Kit supports both on-device and cloud-based data models, providing easy offline functionality when needed.

Why Untappd?

It's all fine and good to choose frameworks and services that help you achieve a technical goal, but apps are nothing these days without data. So when we look at retrieving beer-related data, we have a few choices in RateBeer, BeerAdvocate, and Untappd.

After a little analysis, Untappd was the obvious choice. Not only is it a wildly popular beer-review service, but it has a well documented (and free) API. 🍻

How the App Works

For this app we are going to leverage ML Kit's text recognition APIs. Yes, it's kind of like OCR, but way cooler because we get to say we are using machine learning!

If you're interested in the full source code of the app I'm about to show you, you may consult this GitHub repo.

The simplified flow of the app is:

Take a Picture --> ML Kit Processes Picture --> Get Beer Info from Untappd

Let's first see it in action on iOS:

hoppy example

Note that beer recognition heavily depends on the clarity and styling of the text on a given menu! Don't be surprised if you only find a 30-40% match on most menus...


Take a Picture with NativeScript

We're going to take a picture of a beer menu using the NativeScript camera module. You can think of NativeScript modules as abstractions over native iOS and Android APIs.

    import { Image } from "ui/image";

        then((imageAsset) => {
            console.log("Result is an image asset instance");
            var image = new Image();
            image.src = imageAsset;
        }).catch((err) => {
            console.log("Error -> " + err.message);
Enter fullscreen mode Exit fullscreen mode

Call the ML Kit Text Recognition APIs

Once we have captured an image, we need to send the image file to Firebase for text recognition analysis. Here is how we handle this in the app:

    export function takePicture() {
        .then(imageAsset => {

          const source = new imageSourceModule.ImageSource();

          source.fromAsset(imageAsset).then(imageSource => {
                image: imageSource,
                modelType: 'latest', // default "stable"
                maxResults: 10 // default 10
              .then(function(result) {
                // `result` is a line-by-line output of text
                // in the app we use some regex here to filter out prices and other invalid responses

                let beers = result.split('\n');

                for (let i = 0; i < beers.length; i++) {
                  // hit the Untappd API here!
              .catch(function(errorMessage) {
                // log the ML Kit error to Crashlytics
                console.log('MLKIT ERROR: ' + errorMessage);
                  message: 'Error with ML Kit: ' + errorMessage,
                  showInConsole: true
        .catch(function(err) {
          // show a native alert to the user
            title: 'Camera Issue!',
            message: err.message,
            okButtonText: 'OK',
            cancelable: false
Enter fullscreen mode Exit fullscreen mode

A keen eye will notice that we are using Firebase Crashlytics for error logging as well. Crashlytics is also supported in the NativeScript Firebase plugin.

As previously noted, ML Kit allows us to use on-device or cloud-based text recognition. In this example we are using recognizeTextCloud, but you can swap in recognizeTextOnDevice as well. You can even use NativeScript's connectivity module to determine whether or not your user is online, and thus use the appropriate ML Kit text recognition API:

    const connectionType = connectivityModule.getConnectionType();

    switch (connectionType) {
        case connectivityModule.connectionType.none:
            // Use recognizeTextOnDevice
        case connectivityModule.connectionType.wifi ||
            // Use recognizeTextCloud
Enter fullscreen mode Exit fullscreen mode

Along with text, ML Kit also returns bounding polygon data, showing you where on the image it identified the text. Potential augmented reality application anyone!?

Call the Untappd APIs

In order to avoid a ridiculous amount of code above, I did put a comment placeholder in where the magic of hitting the Untappd API occurs. Those API calls are relatively simple.

First, we need to perform a beer search to see if the text returned from ML Kit is actually a beer name:" + beer + "&client_id=<CLIENT_ID>&client_secret=<CLIENT_SECRET>&access_token=" + access_token
Enter fullscreen mode Exit fullscreen mode

For clarification, beer is the beer name we want to check, client_id and client_secret are the keys you'll get after you register your app with Untappd, and access_token is the auth token provided when you log your user into Untappd.

NOTE: You have the option of not authenticating your end users, but then API calls are restricted to 100/app/hour by Untappd.

If the API call returns a result, we then have to make another API call to retrieve more detailed information about the beer:" + + "?compact=true&client_id=<CLIENT_ID>&client_secret=<CLIENT_SECRET>&access_token=" + access_token
Enter fullscreen mode Exit fullscreen mode

Again, in this case is just the beer id returned from our first API call.

This will return a nice JSON-formatted resultset for us to populate our UI (below being a heavily-condensed version for readibility).

Spotted Cow is an honored favorite of Wisconsin microbrews! πŸΊπŸ„

         "beer_name":"Spotted Cow",
         "beer_description":"You know you're in Wisconsin when you see the Spotted Cow.",
         "beer_style":"Saison \/ Farmhouse Ale",
        "brewery": {
          "brewery_id": 907,
          "brewery_name": "New Glarus Brewing Company",
          "brewery_type": "Micro Brewery",
          "brewery_label": "",
          "brewery_description": "Welcome to our quaint little brewery nestled on the outskirts of New Glarus, Wisconsin."
Enter fullscreen mode Exit fullscreen mode

We then bind the resultset to a NativeScript ListView to provide a native scrolling list (presented here without class attributes to improve readibility).

    <ListView items="{{ myBeers }}" itemTap="beerTap">
            <GridLayout rows="auto" columns="auto, *, auto">
                <Image row="0" col="0" src="{{ image }}" />
                <StackLayout row="0" col="1" >
                    <Label text="{{ name }}" />
                    <Label text="{{ brewery }}" />
                    <Label text="{{ style }}" />
                <StackLayout row="0" col="2">
                    <StackLayout orientation="horizontal">
                        <Label text="{{ score }}" />
                        <Label text=" Avg" />
                    <StackLayout orientation="horizontal">
                        <!-- rating icons provided via background-image CSS property -->
                        <StackLayout class="{{ score_img_1 }}"></StackLayout>
                        <StackLayout class="{{ score_img_2 }}"></StackLayout>
                        <StackLayout class="{{ score_img_3 }}"></StackLayout>
                        <StackLayout class="{{ score_img_4 }}"></StackLayout>
                        <StackLayout class="{{ score_img_5 }}"></StackLayout>
Enter fullscreen mode Exit fullscreen mode

listview screen

A NativeScript ListView is an abstraction on top of Android's android.widget.ListView and iOS's UITableView

You'll notice the beerTap method executed on itemTap. This will navigate us to a new view that shows more details about the selected beer:

    <StackLayout orientation="horizontal">
        <Image src="{{ image }}" />
            <Label text="{{ name }}" textWrap="true" />
            <Label text="{{ brewery }}" textWrap="true"  />
            <Label text="{{ style }}" textWrap="true"  />
    <StackLayout orientation="horizontal">
        <!-- rating icons provided via background-image CSS property -->
        <StackLayout class="{{ score_img_1 }}"></StackLayout>
        <StackLayout class="{{ score_img_2 }}"></StackLayout>
        <StackLayout class="{{ score_img_3 }}"></StackLayout>
        <StackLayout class="{{ score_img_4 }}"></StackLayout>
        <StackLayout class="{{ score_img_5 }}"></StackLayout>
        <Label text="{{ score }}" />
        <Label text=" Avg" />
    <Label text="{{ desc }}" textWrap="true" />
    <StackLayout orientation="horizontal">
        <!-- progress bars provided by NativeScript UI -->
        <circProgAbv:CircularProgressView id="cpAbv" />
        <circProgIbu:CircularProgressView id="cpIbu" />
    <Button text="Open in Untappd" tap="openUntappd" />
Enter fullscreen mode Exit fullscreen mode

hoppy detail screen


The power of leveraging Firebase services with NativeScript and the NativeScript-Firebase plugin can't be understated. You can create dynamic, engaging mobile apps for both iOS and Android from one codebase; saving time, money, and a load of headaches.

You can get started with NativeScript (without installing anything on your local machine) by accessing the NativeScript Playground. Setting up a new Firebase account takes about a minute and has very generous free offerings.

The app you see here (Hoppy) is completely free and available on the iOS App Store and Google Play.

Cheers! 🍻

Discussion (2)

andreslopezrm profile image

Fantastic article, I like NativeScript !!!!