Empowering modern apps using Android ML Kit, TensorFlow, AutoML & On-Device AI.

1. Aba 2025 Building Android Applications That Can See, Hear & Understand Empowering modern apps using Android ML Kit, TensorFlow, AutoML & On-Device AI." Iyke Emmanuel Android dev, FCMB
2. 02 “The era of application is over. Welcome to the era of intelligent applications”
3. Aba Emmanuel Iyke EmmanuelDav_K EmmanuelDav Emmanueldav Android Developer, FCMB
4. What today’s session covers: • ML foundations • ML on mobile • Pretrained models vs custom models • • • • TensorFlow Lite, AutoML, Firebase Kotlin integration Vision, Audio, NLP use cases Practical examples + libraries
5. 05 What is Machine Learning? ML is a field of AI where systems learn patterns from data instead of explicit programming. They “predict”, “classify”, “detect”, or “understand” from examples
6. 06 “Machine Learning is the most dominant and modern way to integrate AI into your software”
7. 07 Types of Machine Learning Supervised Learning is the process you use when you know exactly what the output should be for a given input. It's about learning a function that maps an input to an output based on example input-output pairs
8. • Unsupervised Learning is used when you don't have labels. The model is asked to explore the data and find hidden structures, groupings, or patterns that might not be obvious • Reinforced Learning: agent (the model) that interacts with an environment. It learns the best sequence of actions to maximize a cumulative reward. It learns through trial and error, like training a dog.
9. Application of Machine Learning: • Smart camera apps • Speech transcription • Chat summarization • Fraud detection • Health/fitness tracking • AR experiences • Document scanning
10. 10 Machine Learning integration approach in mobile development
11. ML Integration approach in App Pre-trained Models: Think of this as the easy button. These are models already trained by Google or others on massive, general datasets. If your problem is common—like scanning a barcode, detecting a face, or translating basic text—you use these. They are fast, accurate, and you simply call an API or an SDK function Custom Models: This is the path for unique challenges. If your problem is specific to your region or domain—like identifying local plant species or a custom ID card—you must train the model yourself. This uses tools like TensorFlow Lite and gives you total control over the data and the outcome, but it requires more work
12. ML Kit integration on Android with pretrained Models Prefers using the camera2 or the camera x Library Consider processing image at low resolution to keep the requirement Waiting for the ML kit result before rendering the image
13. ML Kit Vision Video and Image analysis
14. Text recognition Text recognition allows us to extract text from images, camera or static images Trained to recognize text in over 100 languages including Latin based script and none latin based script such as china or japanese The text recognise segment text into block element, line and symbols
15. dependency { Implementation 'com.google.mlkit:text-recognition:16.0.0-beta6' implementation ‘com.google.mlkit:text-recognition-chinese:16.0.0' }
17. Barcode Scanning Read and recognise most popular barcode like Codebar or QRCode Automatic format detection Run on a device so you can have an internet connection to perform the Scan
18. dependency { implementation 'com.google.mlkit:barcode-scanning:17.2.0' }
19. Digital Ink recognition Recognise handwritten text and drawn images and convert it into unicode text Support 300+ languages and 25+ written system Dynamically download the language asset you want to use
20. Text recognition Text recognition allows us to extract text from images, camera or static images Trained to recognize text in over 100 languages including Latin based script and none latin based script such as china or japanese The text recognise segment text into block element, line and symbols
21. ML Kit integration on Android with custom Models
22. Why do we need custom ML Model • Pretrained ML models can’t solve every problem • You may need domain-specific intelligence • Use cases: 1) Local language classification 2) Nigerian foods recognition 3) Fake ID detection 4) Company-specific predictions
23. Show a step-by-step workflow using a clean diagram.
24. Training Model with tensorflow • TensorFlow is the most widely-used ML framework • Used to train image, audio, NLP, and tabular models • Can run on Google Colab (free GPU!) • Export to TensorFlow Lite easily
25. Find Dataset to train data: KAGGLE • Kaggle offers thousands of free ML datasets • You can find datasets for: 1) Images (faces, foods, plants) 2) Audio (speech, noise, accents) 3) Text (sentiment, languages)
26. THANK YOU