DEV Community: Aayushi Shinde

LazyDesk — Clean Files. Keep Your Brain. 🦥

Aayushi Shinde — Sat, 24 Oct 2026 18:40:40 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady

LazyDesk — Clean Files. Keep Your Brain. 🦥

"Stop sorting files one by one. Let your local AI prepare the plan while you stay in control."

What I Built

LazyDesk is an open-source, local-first file cleanup agent engineered to eliminate the manual chore of organizing messy directories like Downloads or Desktop without sending a single file name or document to cloud APIs.

Unlike basic file-organizer scripts that blindly sort strictly by .ext (dumping code snippets into documents or bank receipts into random pictures), and unlike unpredictable autonomous agent scripts that execute raw shell commands, LazyDesk is built on the "Slow and Steady" principle:

Metadata-Only AI Planning: LazyDesk scans an approved directory, extracts file metadata (filename, byte size, extension), and streams this structure to a local open-weight model (Qwen2.5:3b-instruct or 1.5b-instruct via Ollama).
Contextual Categorization: The model categorizes files into purpose-built buckets (Career and Resumes, Study Materials, Financial & Receipts, Source Code, Archives, Installers, Media) with confidence scores (0–100%), risk ratings, and plain-English rationales.
Interactive Human-in-the-Loop Review: Users can inspect every proposed move in a React 19 dashboard, edit destination folders inline, filter by risk level, and uncheck files.
Collision-Immune Execution & Dry-Run: Performs an in-memory dry-run before touching disk. LazyDesk refuses to overwrite existing files, skipping collisions automatically.
Cryptographic Manifest & 1-Click Rollback: Every executed batch is saved into .lazydesk_manifest.json with SHA-256 hashes and timestamps. A single click on "Undo Batch" verifies the destination and restores every file to its exact origin.
Bitwise Duplicate Detective: Uses chunked 64 KB SHA-256 streaming hashes (independent of filename) to group identical files and compute exact recoverable bytes without double-counting.

Who gets to be lazier?

Developers with hundreds of unzipped repos, Docker dumps, and .env configs.
Students drowning in lecture slides, syllabi, assignment PDFs, and problem sets.
Freelancers and creators sorting quarterly tax invoices, contracts, and video captures.

Demo

GitHub Repository & Demo Assets: https://github.com/AyushiShinde/LazyDesk

The repository includes complete step-by-step instructions to run LazyDesk locally with one-click PowerShell scripts, generate a 19-file synthetic test sandbox, and test the full scan, AI planning, dry-run simulation, and 1-click rollback cycle.

Code

LazyDesk is completely open-source under the Apache 2.0 license:

GitHub Repository: https://github.com/AyushiShinde/LazyDesk
License: Apache-2.0

Key Architecture Modules:

backend/app/scanner.py: Root boundary containment engine. Validates paths using Path.resolve().relative_to(approved_root) and rejects path traversals (..), symlinks, and system directories (C:\Windows, C:\Program Files).
backend/app/ai_planner.py: Ollama client (http://127.0.0.1:11434) enforcing strict Pydantic v2 schemas (PlanRecommendation). Includes prompt injection sanitization and an offline deterministic keyword classifier fallback.
backend/app/duplicates.py: Memory-efficient 64 KB chunked SHA-256 bitwise hasher for duplicate detection across multi-gigabyte files without memory spikes.
backend/app/operations.py: Dry-run engine, atomic file relocation, collision avoidance, and manifest ledger with rollback verification.
backend/tests/: 19 automated Pytest unit and security tests covering boundary containment, duplicate math, dry-run simulation, and undo conflicts.

How I Built It

LazyDesk was built around an open-weight local stack optimized for everyday consumer laptops (tested on an 8 GB RAM laptop with an NVIDIA RTX 2050 4 GB GPU):

1. Local AI Engine (Ollama + Qwen2.5)

Model: Qwen2.5:3b-instruct (and Qwen2.5:1.5b-instruct for ultra-low memory setups).
Quantization: Q4_K_M (4-bit quantization). The 3B model requires only ~1.9 GB of disk storage and ~2.4 GB of VRAM, running at high token speeds on local hardware.
Structured Schema Enforcement: The prompt feeds an array of file metadata objects and forces a JSON array response conforming to PlanRecommendation(filename, suggested_folder, confidence, reason, risk_level).
Zero-Crash Fallback: If Ollama is offline or unpulled, LazyDesk's RuleBasedPlanner automatically handles classification using keyword heuristics (matching resume tokens, financial indicators, and code extensions) so the application remains 100% functional offline.

2. Defense-in-Depth Security

Because the agent handles real files, safety is enforced in code, never delegated to the LLM:

Prompt Injection Defense: Only filenames, sizes, and extensions are passed to the model. File contents are never read or passed into the prompt, preventing indirect prompt injection from malicious files.
Path Sanitization: Destination paths are validated through sanitize_relative_path(), stripping absolute prefixes, drive letters, and traversal tokens (..).
Collision Protection: Before moving file.pdf into Documents/file.pdf, the engine checks destination.exists(). If occupied, the operation is skipped and recorded as a collision.
Non-Destructive Guarantee: Duplicates are exposed with byte-level SHA-256 proof in the UI, but LazyDesk never deletes files automatically.

Why Does Open Innovation Matter?

Open innovation—specifically open-weight models and open-source inference engines—is the only reason LazyDesk can exist responsibly:

Absolute Privacy for Sensitive Personal Data: A person's Downloads folder contains salary slips, tax returns, medical records, identity scans, and proprietary code. Piping those filenames and directory hierarchies to cloud APIs like OpenAI or Anthropic violates basic privacy hygiene. With open-weight Qwen2.5 running on Ollama, all inference happens strictly over 127.0.0.1.
Zero Recurring Costs or API Keys: Closed APIs require credit cards, monthly subscriptions, token quotas, and active internet connections. Open-weight models are free to run forever, even when working completely offline on a plane or train.
**Reproducible, Auditable Software

RESET-Return to What Matters

Aayushi Shinde — Sun, 11 Oct 2026 05:52:20 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass

What I Built

Re:Set is an open-source, open-weight AI experience designed to help people reclaim their attention from screen addiction and return to tactile offline life—starting with reading.

We built AI that can write code and generate images; we wanted to build an AI whose sole purpose is to help people stop using AI and put down their phones.

Instead of trapping users inside another gamified productivity dashboard, endless chatbot loop, or streak tracker, Re:Set is built on an inverted design principle:

"The better Re:Set works, the less time the user spends inside Re:Set."

How It Gets People Off the Screen and Into the World

Low-Friction Spark: The user shares what is keeping them from reading (e.g., "I used to read a lot, but now I spend my free time scrolling"), and how much time they have (15, 30, or 60 minutes).
Open-Weight Gemma Generation: Google's Gemma 2 synthesizes a tactile, single-action offline ritual featuring a physical location rule, a phone-separation rule, and an attention anchor.
Deliberate Shutdown (The Offline Screen): As soon as the user clicks Begin Re:Set, the application transforms into a serene, distraction-free screen: > "Your reset has started. You don't need this app anymore. Put your phone away. 🌿"
Non-Judgmental Reflection: When they return, Gemma synthesizes their takeaway with zero streaks, zero guilt, and zero gamified penalties—because returning to a habit matters far more than unbroken consistency.
Community Silent Reading Gatherings: Re:Set features a physical Venue QR Gathering System that allows independent bookstores, libraries, and cafés to host silent group reading hours with privacy-preserving attendance verification.

Who Is It For?

Anyone who loves books, nature, and quiet thought but feels their attention hijacked by notification loops and algorithmic feeds.
Independent bookstores, community libraries, and local cafés seeking to host silent reading gatherings without forcing intrusive surveillance onto attendees.

Demo

Interactive Walkthrough Mode: The app includes a "⚡ 1-Minute Judge Walkthrough" banner and a "Simulate 30 Minutes" button so judges can experience the full end-to-end ritual without waiting half an hour or needing a secondary device.

The 10-Step Experience Flow:

The Spark: Express your personal friction and select a duration.
Gemma Challenge: Receive your customized location rule and tactile anchor.
Offline Transition: App steps out of the way; phone goes face-down.
Community Venue Check-In: Scan the physical START QR at your local venue.
Show Up, Not Tracked: Venue displays real-time collective presence (27 checked in).
Session Completion: Scan the COMPLETE QR upon finishing.
Honest Reflection: Share a private takeaway without surveillance.
Gemma Synthesis: Non-judgmental distillation celebrating presence.
The Passport: View cumulative offline time returned to real life (e.g., 2h 15m returned).
The Exit: "The best AI interaction might be the one that ends with you closing the app."

Code

AayushiShinde / RESET

Re:Set

Return to what matters.
An AI that helps you use your phone less.

Built for Hacktoberfest 2026 Open-Source AI Challenge — Week 1: Touch Grass.

Core Product Principle

"The better Re:Set works, the less time the user spends inside Re:Set."

Re:Set is not another AI chatbot, productivity dashboard, habit tracker, streak app, or wellness assistant. It is a bridge between artificial intelligence and real-world life.

1. The Problem

People increasingly spend their free time trapped in infinite feeds and notification loops, even when they genuinely want to return to offline activities they love: reading physical books, walking in nature, moving their bodies, and gathering in person.

Traditional apps claim to solve this by adding more screen time: demanding daily check-ins, punishing broken streaks, sending artificial notifications, and treating attention as an engagement metric.

2. The Idea

Re:Set uses an open-weight AI model (Gemma 2) to create…

View on GitHub

Deployed on

https://reset-touch-grass.onrender.com)

Re:Set is 100% open-source under the Apache-2.0 license.

Architecture Highlights

Client: React 18, Vite, Lucide Icons, QRCode.react, crafted with an editorial, paper-inspired aesthetic and zero telemetry scripts.
Server: Node.js, Express, Mongoose.
AI Engine: Decoupled AIService routing to Google Gemma 2 (via Hugging Face API or local Ollama) with a deterministic offline fallback.
Database: MongoDB Atlas for community gatherings and offline passports (with automatic in-memory fallback for local environments).
Observability: Custom Sentry agent tracing around generateReset(), completeSession(), and analyzeReflection().

How I Built It

1. Open-Weight AI Architecture (Google Gemma 2)

Re:Set is powered by Google's open-weight model family (Gemma 2 9B / 2B). We built a modular provider interface:


text
       ┌──────────────┐
       │  AIService   │
       └──────┬───────┘
              │
       ┌──────┴───────┐
       ▼              ▼
┌──────────────┐ ┌──────────────┐
│GemmaProvider │ │ Gemma-Local  │
│ (Cloud Host) │ │(Ollama:11434)│
└──────────────┘ └──────────────┘

![ ](https://dev-to-uploads.s3.us-east-2.amazonaws.com/uploads/articles/6r44xcvgkntidykh3s8y.png)

MaaMode: "Don’t Make Mom Learn the Phone. Make the Phone Learn Mom" ❤️

Aayushi Shinde — Mon, 05 Oct 2026 06:19:37 +0000

This is a submission for the Hacktoberfest Weekend Challenge: Build for my mom

## What I Built

"Don't make Mom learn the phone. Make the phone learn Mom."

Every time I visited my mother, I noticed the same pattern: she was anxious about using her smartphone. Whether it was decoding a cryptic English SMS from the bank, worrying she might accidentally send money to the wrong person, trying to send her live location on WhatsApp when visiting relatives, or booking an auto on Uber, she felt overwhelmed by tiny icons, technical jargon, and English-first interfaces.

She would constantly ask: "Is this safe to click?", "Did I delete something?", or "Can you just do this for me?"

I built MaaMode (माझं Mode / MaaMate) specifically for my Mom (आई).

MaaMode is an empathetic, bilingual AI smartphone bridge and interactive personal tutor designed to remove smartphone anxiety for Indian mothers and elderly family members.

Core Features Built with Love for Mom:

Giant Tactile Vernacular UI: High-contrast, tactile 2x2 cards with Marathi (मराठी), Hindi (हिंदी), and English modes. No confusing nested menus or accidental micro-clicks.
Warm Vernacular Voice Assistant: Mom doesn't have to read tiny fonts. She can simply tap the giant microphone and speak naturally ("लोकेशन कशी पाठवायची?" or "पैसे कसे पाठवायचे?").
Zero-Risk Guided Practice Sandboxes: Realistic smartphone simulation environments where Mom can safely practice without fear of breaking anything:
- 📸 WhatsApp Photo & Camera sharing
- 📄 Document / Aadhaar PDF sharing
- 📍 Live GPS Location sharing
- 💸 UPI Payment & QR scanning safety
- 🛺 Uber Auto booking step-by-step
- 🌟 WhatsApp Status with Captions
Smart SMS & Message Simplifier (Open-Weights AI): Translates confusing English bank warnings, parcel delivery OTPs, doctor appointments, and electricity bills into calming, straightforward Marathi/Hindi summaries with 3 pre-vetted, respectful one-tap reply options.
Dual Architecture (Web PWA + Native Android Overlay):
- A 1-tap installable Mobile Progressive Web App (PWA) that runs offline.
- A companion Native Android Studio App that uses AccessibilityService and SYSTEM_ALERT_WINDOW to project a glowing gold spotlight ring directly over real apps (like WhatsApp and Uber) to guide her fingers live.

## Demo

Live Repository: github.com/AayushiShinde/MaaMode
Render Deployment: Ready-to-deploy 1-click blueprint via render.yaml

Key User Flows in Action:

Vernacular Voice Guidance: Tap the microphone and hear warm, natural maternal Marathi responses powered by ElevenLabs Multilingual v2.
Message Simplification: Paste a formal English medical or delivery alert like "Doctor's appointment is confirmed for tomorrow at 11:30 AM. Please bring blood reports" — MaaMode instantly summarizes: "उद्या सकाळी ११:३० वाजता डॉक्टरांची भेट आहे. रक्ताचे रिपोर्ट सोबत ठेवायचे आहेत." and offers safe replies like "हो, मी वेळेवर पोहोचेन".
Interactive Sandbox: Step-by-step glowing spotlight rings guiding Mom through attaching documents, verifying UPI payee names, and booking an auto safely.

Code

AayushiShinde / MaaMode

MaaMode (माझं Mode / MaaMate) ❤️

"Don't make Mom learn the phone. Make the phone learn Mom."
Built for Mom for DEV Hacktoberfest Weekend 2026: "Build for a Friend"

📁 Project Directory Overview

This master folder (C:\Users\Ayushi_Shinde\MaaMode) contains both parts of the solution:

1. MaaMode-Web/ (Full PWA Web App + Cloud Backend)

What it is: Mobile Progressive Web App with offline service worker, interactive smartphone sandbox, and bilingual voice engine.
Includes:

ender.yaml: 1-click Render blueprint deployment (**Best Use of Render - **).

server.js: Node.js/Express server with secure ElevenLabs TTS proxy.
�oice-engine.js: Integrated with ElevenLabs Multilingual v2 maternal voice (**Best Use of ElevenLabs - **).
�i-engine.js: Google Gemma 2 local intent classifier & message simplifier (**Best Use of Gemma - **).

utorial-sandbox.js: Complete interactive guided tutorials (WhatsApp Photo, Document/Aadhaar PDF, Live GPS Location, UPI Payment, WhatsApp Status with Caption, Uber Auto Booking)

sw.js & manifest.json: Offline PWA installation on…

View on GitHub

### Architecture Breakdown:

MaaMode-Web/:
- ai-engine.js: Open-weights AI inference engine with a strict intent whitelist validator and offline semantic fallback.
- voice-engine.js: Dual-layer voice synthesis with ElevenLabs Multilingual v2 streaming and on-device Web Speech API resilience.
- tutorial-sandbox.js: State-machine driven smartphone sandboxes with simulated WhatsApp chat, document pickers, and UPI gateways.
- sw.js & manifest.json: Offline PWA service worker with 1-tap home-screen installation.
MaaMode-Android/:
- MaaModeOverlayService.kt: Native Android floating window service rendering dynamic spotlight rings.
- MaaModeAccessibilityService.kt: Live screen-tree inspector for com.whatsapp and com.ubercab.
server.js & render.yaml:
- Node.js/Express backend with healthcheck endpoints and zero-downtime Render deployment.

How I Built It

MaaMode is built around a Controlled Action Architecture combining open-source AI, maternal voice synthesis, and native mobile accessibility:

DEV Community: Aayushi Shinde

LazyDesk — Clean Files. Keep Your Brain. 🦥

What I Built

Demo

Code

Key Architecture Modules:

How I Built It

1. Local AI Engine (Ollama + Qwen2.5)

2. Defense-in-Depth Security

Why Does Open Innovation Matter?

RESET-Return to What Matters

What I Built

How It Gets People Off the Screen and Into the World

Who Is It For?

Demo

The 10-Step Experience Flow:

Code

AayushiShinde / RESET

Re:Set

Core Product Principle

1. The Problem

2. The Idea

Deployed on

Architecture Highlights

How I Built It

1. Open-Weight AI Architecture (Google Gemma 2)

MaaMode: "Don’t Make Mom Learn the Phone. Make the Phone Learn Mom" ❤️

Core Features Built with Love for Mom:

Key User Flows in Action:

Code

AayushiShinde / MaaMode

MaaMode (माझं Mode / MaaMate) ❤️

📁 Project Directory Overview

1. MaaMode-Web/ (Full PWA Web App + Cloud Backend)

Includes:

How I Built It