Project Overview
De-Droid started from a simple frustration: cleaning Android bloatware is often messy, risky, and full of guesswork. I wanted a tool that feels practical for real people, not just power users running random shell commands from old forum threads.
So I built De-Droid as a local-first desktop app that lets users inspect packages, understand risk, and take safe actions with confidence.
What Makes It Useful
- Safety-first debloating: Package actions are guided by risk labels like
RECOMMENDED,ADVANCED,EXPERT, andUNSAFE. - Human-friendly AI/ML layer: Users can see confidence scores and top factors behind model decisions instead of black-box output.
- Practical recovery flow: Backups and action history make it easier to undo mistakes.
- Everyday usability: Search, filter, and batch actions reduce repetitive manual work.
- Open-source alternatives: The app suggests replacement options where applicable.
AI/ML Side (In Plain Words)
The project includes an optional FastAPI model service and a training pipeline that converts package data into safety predictions used by the desktop app.
The goal is not to blindly automate removals. The goal is to support better decisions:
- classify package safety levels,
- show confidence,
- expose model reasoning,
- and apply conservative safety gates for risky cases.
That blend of machine learning plus guardrails is what gives De-Droid its core value.
Tech Stack
- Desktop App: Electron, React, TypeScript
- Local Runtime: ADB integration, JSON + SQLite for local data
- Model/API Layer: Python, scikit-learn pipeline, FastAPI
- Workflow: Prediction export from model pipeline to desktop runtime
Outcome
De-Droid became a full product-style build where I worked across desktop UX, local device tooling, and AI/ML-backed decision support. It demonstrates how I approach real-world developer tools: make them powerful, but also understandable and safe for normal users.
Screenshots
