androidengineers.Book a session

LEARN WITH A CLEAR DIRECTION

Gemma on Android with PocketStories — Preview

Learn app-managed on-device AI through PocketStories: model downloads, CameraX, LiteRT-LM, Compose, context, cancellation, Room and device diagnosis.

IntermediateSelf-paced13 modules13 available lessons2 practical labs · 1 capstone

Your path, one skill at a time.

A complete 13-unit learning path around the published PocketStories preview. Each unit connects explanation, source and an exercise; the codelab includes a red/green context-policy extension and an independent capstone. Software checks are verified; successful end-to-end Gemma generation remains under validation. The tested emulator encounters a GPU buffer limit.

Work through the 13 units in order. Keep an evidence log. Complete software exercises even when native inference is blocked; report the live-model track separately.

What you’ll learn

  • Acquire and verify a resumable app-private model
  • Prepare deliberate camera input and trace MVVM state
  • Explain LiteRT engine ownership, streaming and cancellation
  • Implement and test bounded accepted-scene context
  • Persist drafts and evaluate offline behavior with honest device evidence

YOUR PROGRESSION

Know what you can build at every stage.

Complete the labs and keep the evidence. Use each checkpoint to decide whether to move forward or revisit a skill.

  1. 01

    Units 1–3

    Prepare

    Reproduce the baseline and acquire the correct model

    Evidence: Build log, artifact manifest and downloader tests

    Go to this stage’s final module →
  2. 02

    Units 4–7

    Connect

    Trace camera input, state, runtime and cancellation

    Evidence: Camera recovery check and resource-ownership diagram

    Go to this stage’s final module →
  3. 03

    Units 8–10

    Extend

    Bound context, persist stories and diagnose failures

    Evidence: Red/green context tests and device failure report

    Go to this stage’s final module →
  4. 04

    Units 11–13

    Evaluate

    Complete a measured evidence matrix and independent capstone

    Evidence: Capstone diff, regression tests and honest assessment

    Go to this stage’s final module →

THE CURRICULUM

Build your knowledge, step by step.

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

01MODULE 01PocketStories study guide1 lessons · Self-paced
  • PocketStories study guide
  • Source walkthrough
  • Practical verification

Build the pinned app and separate software evidence from inference evidence.

  1. 01PocketStories study guideLesson · Self-paced

Ready to move on when: Build the pinned app and separate software evidence from inference evidence.

02MODULE 02Models and device fit1 lessons · Self-paced
  • Models and device fit
  • Source walkthrough
  • Practical verification

Choose a vision-capable artifact and explain storage, memory and backend constraints.

  1. 01Models and device fitLesson · Self-paced

Ready to move on when: Choose a vision-capable artifact and explain storage, memory and backend constraints.

03MODULE 03Verified model downloads1 lessons · Self-paced
  • Verified model downloads
  • Source walkthrough
  • Practical verification

Trace HTTP resume, cancellation and verified promotion into private storage.

  1. 01Verified model downloadsLesson · Self-paced

Ready to move on when: Trace HTTP resume, cancellation and verified promotion into private storage.

04MODULE 04Camera to model input1 lessons · Self-paced
  • Camera to model input
  • Source walkthrough
  • Practical verification

Capture one intentional frame and prepare a bounded private image.

  1. 01Camera to model inputLesson · Self-paced

Ready to move on when: Capture one intentional frame and prepare a bounded private image.

05MODULE 05Compose state and boundaries1 lessons · Self-paced
  • Compose state and boundaries
  • Source walkthrough
  • Practical verification

Trace an action through immutable state and injected data services.

  1. 01Compose state and boundariesLesson · Self-paced

Ready to move on when: Trace an action through immutable state and injected data services.

06MODULE 06LiteRT runtime and streaming1 lessons · Self-paced
  • LiteRT runtime and streaming
  • Source walkthrough
  • Practical verification

Understand engine setup, multimodal messages and explicit native resource ownership.

  1. 01LiteRT runtime and streamingLesson · Self-paced

Ready to move on when: Understand engine setup, multimodal messages and explicit native resource ownership.

07MODULE 07Cancellation and recovery1 lessons · Self-paced
  • Cancellation and recovery
  • Source walkthrough
  • Practical verification

Distinguish a cancelled coroutine from completed native cleanup.

  1. 01Cancellation and recoveryLesson · Self-paced

Ready to move on when: Distinguish a cancelled coroutine from completed native cleanup.

08MODULE 08Bounded story context1 lessons · Self-paced
  • Bounded story context
  • Source walkthrough
  • Practical verification

Implement a scene-count limit with a failing test and a verified fix.

  1. 01Bounded story contextLab · Self-paced

Ready to move on when: Implement a scene-count limit with a failing test and a verified fix.

09MODULE 09Room stories and drafts1 lessons · Self-paced
  • Room stories and drafts
  • Source walkthrough
  • Practical verification

Persist accepted stories and recover drafts without persisting native sessions.

  1. 01Room stories and draftsLesson · Self-paced

Ready to move on when: Persist accepted stories and recover drafts without persisting native sessions.

10MODULE 10Diagnose device failures1 lessons · Self-paced
  • Diagnose device failures
  • Source walkthrough
  • Practical verification

Classify acquisition, native-loading and inference errors without unnecessary downloads.

  1. 01Diagnose device failuresLesson · Self-paced

Ready to move on when: Classify acquisition, native-loading and inference errors without unnecessary downloads.

11MODULE 11Evaluate offline generation1 lessons · Self-paced
  • Evaluate offline generation
  • Source walkthrough
  • Practical verification

Measure actual output separately from deterministic app tests.

  1. 01Evaluate offline generationLesson · Self-paced

Ready to move on when: Measure actual output separately from deterministic app tests.

12MODULE 12PocketStories capstone1 lessons · Self-paced
  • PocketStories capstone
  • Source walkthrough
  • Practical verification

Design and implement persisted character facts with strict context boundaries.

  1. 01PocketStories capstoneLab · Self-paced

Ready to move on when: Design and implement persisted character facts with strict context boundaries.

13MODULE 13PocketStories final assessment1 lessons · Self-paced
  • PocketStories final assessment
  • Source walkthrough
  • Practical verification

Explain the architecture and present reproducible learning evidence.

  1. 01PocketStories final assessmentLesson · Self-paced

Ready to move on when: Explain the architecture and present reproducible learning evidence.

PROVE YOUR SKILLS

A portfolio you can build, explain, and defend.

  • Reproduce the pinned baseline and classify test evidence
  • Pass the new context-policy tests after observing them fail
  • Explain download integrity, image preparation and native cleanup
  • Submit the character-facts capstone and explicit device verification matrix

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 →