This is a submission for the Gemma 4 Challenge: Build with Gemma 4
Blossom — on-device AI companion for student wellbeing
In India, getting into top institutes is extremely competitive. There are very limited seats, and the pressure builds up over time. For most students, this leads to constant stress, burnout, and in some cases, suicides.
That environment is the reason we started building Blossom.
Blossom is still under development, but the goal is simple: to give students a private, always-available companion that helps them manage both studies and mental wellbeing.
Team
What We Built
Blossom is an Android chat app designed around student life.
It combines an on-device AI assistant with study planning and wellbeing support. The assistant runs locally using Gemma, which means it works offline, responds quickly, and keeps conversations private.
The assistant can also help users create and schedule routines and tasks directly through conversation, making it easier to turn plans into actionable steps.
Alongside productivity, Blossom focuses on mental wellbeing by offering lightweight, accessible support during stressful moments.
It is not meant to replace real support systems, but to be something that is available anytime a student needs guidance, structure, or just a small check-in.
Demo
Code
GitHub:
Blossom
Blossom is an Android chat app focused on student mental wellbeing. It runs an on-device Gemma assistant, provides a routines manager, and supports offline Kokoro TTS for assistant replies.
This README: quick setup, build/run instructions, where to find Kokoro assets, and testing commands.
Requirements
- JDK 11+ (matching Android Gradle Plugin requirements).
- Android Studio (Arctic Fox or newer recommended) with Android SDK for API 26+.
- Gradle wrapper is included; prefer using the wrapper (
./gradlew/gradlew.bat).
Quick start
- Open the project root in Android Studio (the folder containing
settings.gradle.kts). - Let Android Studio sync Gradle and download any missing SDK components.
- Run the app on a device or emulator with API level 26 or higher.
Build from command line
From the project root, build the debug APK with:
./gradlew :app:assembleDebug
Install to a connected device/emulator:
./gradlew :app:installDebug
Clean build outputs:
./gradlew clean
On Windows use gradlew.bat instead…
Current features
On-device Gemma study assistant
Blossom runs a local Gemma model directly on the device to help students plan study sessions, stay focused, and get supportive wellbeing guidance. It can also create and schedule routines and tasks through natural conversation, along with supporting voice input and media attachments.
Study plan manager
Create, edit, activate, and delete study plans and tasks. The app includes a dedicated schedule view, highlights today’s tasks, and keeps reminders updated for the active plan.
Mental wellbeing support
Provides stress-management suggestions and simple coping strategies alongside study assistance.
Offline Kokoro text-to-speech
Assistant replies can be spoken out loud fully offline, with configurable voices and one-tap playback directly from chat bubbles.
Everyday usability
Copy any message, listen to assistant responses, reset conversations, and use quick-start prompts for common student needs.
Reliable reminders
Uses local notifications, exact alarms and TTS for plan and task reminders, with automatic rescheduling after device restarts so active plans stay on track.
How We Used Gemma 4
Blossom is powered by Gemma 4 E2B (instruction-tuned) running fully on-device using LiteRT.
The model is responsible for:
- Generating study plans and structuring tasks
- Providing contextual suggestions to stay focused
- Offering basic mental wellbeing guidance
- Interpreting user intent to create and schedule routines
An on-device approach was chosen to prioritize:
- Privacy — conversations never leave the device
- Offline availability — useful for students with limited connectivity
- Low latency — faster and more natural interactions
Given the constraints of mobile devices, a smaller and efficient Gemma variant (such as E2B/E4B) is the most practical fit. It balances performance and resource usage while still delivering meaningful assistance.
Why Gemma 4 E2B?
We intentionally chose Gemma 4 E2B because Blossom is designed as a mobile-first, offline-capable application.
Why LiteRT?
We used LiteRT to deploy the model because it enables:
- Efficient on-device inference
- Reduced memory footprint
- Better compatibility with Android environments
This combination allows Blossom to remain accessible even on mid-range devices, which is critical for the target user base.
Why on-device?
A key design choice in Blossom is running the AI locally.
This approach improves privacy, which is important for sensitive conversations around stress and mental health. It also allows the app to function without internet access and reduces response latency.
Future plans
Voice-first interaction
Making conversations feel more natural, closer to talking with a companion rather than typing to a system.
Multilingual support
Supporting multiple Indian languages so the app is accessible to more students.
Model and storage optimization
Reducing size and improving performance so it works well on a wider range of devices.
Hybrid local and cloud option
Giving users the choice between fully offline usage or cloud-based models for more advanced responses.
Stronger preventive wellbeing features
Exploring ways to detect patterns, identify early signs of stress, and provide more proactive support.
Why this matters
This is not just a technical project.
If something like Blossom can help even a small number of students feel more supported, more structured, or slightly less overwhelmed, then it is worth building.
Feedback
The project is still evolving, and feedback is genuinely welcome. Suggestions on features, usability, or ways to better support students would be really helpful.
Top comments (0)