androidengineers.Book a session

LEARN WITH A CLEAR DIRECTION

On-device AI on Android

Understand local inference, LiteRT-LM model delivery, quantization, memory budgets and hardware benchmarking.

Intermediate to AdvancedSelf-paced7 modules13 available lessons

Your path, one skill at a time.

Understand local inference, LiteRT-LM model delivery, quantization, memory budgets and hardware benchmarking. Follow six teaching modules, each with a practical assignment, then combine the skills in the final project. Start with the introduction even if the tools are familiar: it defines the scope and boundaries of the path.

Self-paced. Complete each exercise and use its checkpoint before moving to the next module. Revisit any prerequisite you cannot yet demonstrate.

What you’ll learn

  • Verify model integrity and compatibility before loading
  • Benchmark cold start, warm inference and sustained use
  • Keep input local and recover from insufficient storage

THE CURRICULUM

Build your knowledge, step by step.

13 lessons across 7 modules. Open a module to explore its lessons and practical work.

01MODULE 01What on-device inference changes2 lessons · Self-paced
  • Runtime selection
  • Model artifacts
  • Offline promise

Understand the deployment responsibilities that move into the app.

  1. 01What on-device inference changesLesson · Self-paced
  2. 02Practice: What on-device inference changesLab · Self-paced

Ready to move on when: You can name the assets and runtime assumptions required for offline operation.

02MODULE 02Model selection and quantization2 lessons · Self-paced
  • Parameters and memory
  • Quantization tradeoffs
  • License and provenance

Choose a model that fits the task and the device budget.

  1. 01Model selection and quantizationLesson · Self-paced
  2. 02Practice: Model selection and quantizationLab · Self-paced

Ready to move on when: The selected artifact fits both the device budget and the task acceptance criteria.

03MODULE 03Delivery, verification and storage2 lessons · Self-paced
  • Download lifecycle
  • Integrity and atomic activation
  • Cache and deletion

Treat model assets as versioned application dependencies.

  1. 01Delivery, verification and storageLesson · Self-paced
  2. 02Practice: Delivery, verification and storageLab · Self-paced

Ready to move on when: Only a fully verified package becomes the active model.

04MODULE 04Runtime integration and inference ownership2 lessons · Self-paced
  • Initialization scope
  • Execution and cancellation
  • Concurrency limits

Keep expensive local work away from the UI thread.

  1. 01Runtime integration and inference ownershipLesson · Self-paced
  2. 02Practice: Runtime integration and inference ownershipLab · Self-paced

Ready to move on when: Your UI remains responsive and ownership of native resources is unambiguous.

05MODULE 05Hardware acceleration and context costs2 lessons · Self-paced
  • CPU and accelerators
  • Context growth
  • Thermal and battery limits

Benchmark the actual deployment configuration.

  1. 01Hardware acceleration and context costsLesson · Self-paced
  2. 02Practice: Hardware acceleration and context costsLab · Self-paced

Ready to move on when: Your performance claims include device, backend, model and workload.

06MODULE 06Quality evaluation and model upgrades2 lessons · Self-paced
  • Task datasets
  • Artifact compatibility
  • Staged activation

Ship model changes with the discipline of software releases.

  1. 01Quality evaluation and model upgradesLesson · Self-paced
  2. 02Practice: Quality evaluation and model upgradesLab · Self-paced

Ready to move on when: You can roll back a model independently from ordinary user documents.

07MODULE 07Capstone and portfolio review1 lessons · Self-paced
  • Verify model integrity and compatibility before loading
  • Benchmark cold start, warm inference and sustained use
  • Keep input local and recover from insufficient storage

Build an offline document classifier with a versioned model package, measured resource use and recovery from corrupted downloads.

  1. 01Capstone: On-device AI on AndroidProject · Self-paced

Ready to move on when: Demonstrate every acceptance criterion and explain the tradeoffs without relying on the lesson text.

PROVE YOUR SKILLS

A portfolio you can build, explain, and defend.

  • Verify model integrity and compatibility before loading
  • Benchmark cold start, warm inference and sustained use
  • Keep input local and recover from insufficient storage

Use the final review rubric to identify gaps. Lesson completion tracks study progress; readiness comes from independently demonstrating the work.

Open the portfolio review rubric →