LazyDesk AI: Channeling Your Inner Sloth with Local Google Gemma 2

G Surya Prakash — Wed, 21 Oct 2026 16:24:01 +0000

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

What I Built

Rushing through repetitive, mundane computer chores is a waste of mental energy.

For most developers, students, and knowledge workers, our Downloads and Desktop folders are digital wastelands:

Months of unread PDF invoices, contracts, and study notes.
Duplicated presentations and reports named Annual_Report_Final (1).md wasting storage.
Random screenshots and code snippets scattered across directories.

We know we should organize them, extract the action items, and clear out gigabytes of duplicates—but we keep putting it off.

LazyDesk AI rewards your inner sloth. It is a calm, local-first productivity assistant that does the computer chores you keep putting off:

Intelligent Folder Sorting: Scans messy directories in seconds and classifies files into dedicated categories (Documents/, Spreadsheets/, Images/, Code/, Audio_Video/, Archives/) with automatic collision protection (e.g. proposing report_1.pdf instead of overwriting).
Local Document Intelligence: Uses Google Gemma 2 (2B) running 100% on-device to digest multi-page documents, synthesize 2-sentence executive summaries, extract actionable to-do checklists, and detect upcoming calendar deadlines.
Cryptographic Duplicate Detection: Uses two-phase SHA-256 checksum fingerprinting to detect exact identical copies, calculate reclaimable megabytes, and safely archive redundant copies without deleting the original.
Physical Disk Undo: Every move, rename, and archival is recorded in an atomic SQLite WAL-mode log. Clicking Undo Move physically restores the file on your hard drive back to its original location.

Demo

LazyDesk AI features a warm editorial design inspired by modern typography and calm productivity:

Key Workspaces in Action:

Minimalist Front Entrance: A clean, centered welcome portal providing instant access to the workspace.
Overview Dashboard ("A little less to do."): An understated home view showing files scanned, pending organization suggestions, recoverable duplicate storage, and live Gemma engine latency (~5ms).
File Organizer: Visual proposals with live filename search, category dropdown filters, collision warnings, and a safe Dry Run Preview before moving any files.
Document Intelligence: An editorial reader that presents Google Gemma 2's executive summaries, checkboxes, and milestone dates with a 1-click clipboard copy button.
Duplicate Detection: Sets grouped by SHA-256 hash, highlighting the preserved original in soft green and redundant copies in a safe archival queue.
Activity History: Complete audit trail with real filesystem Undo Move action.

(Full visual walkthrough and UI screenshots are available in the GitHub Repository README.)

Code

The complete source code, test suite, and quickstart documentation are available on GitHub:

GSuryaP / LazyDeskAI

LazyDesk AI

Your calm, local-first productivity assistant.
Built with open-weight Google Gemma models for the Hacktoberfest 2026 DEV Challenge: Slow and Steady.

The "Slow and Steady" Story

Rushing through repetitive, mundane computer chores is a waste of mental energy. For most of us, our Downloads and Desktop folders are digital junk drawers overflowing with unread PDFs, duplicated invoices, untitled screenshots, and scattered spreadsheets.

LazyDesk AI rewards your inner sloth. Instead of spending hours manually sorting files and digging for action items:

One click scans your clutter.
Local Google Gemma open-weight models synthesize executive summaries and extract deadlines.
Safe categorization proposals are generated with dry-run previews and collision detection.
Cryptographic SHA-256 fingerprinting finds redundant copies to reclaim valuable gigabytes.
Every single move has a physical disk Undo button.

Your files. Your machine. Zero cloud fees. Zero telemetry.

Features

Warm Editorial Aesthetic: Built with two bespoke, accessibility-tested themes
- Warm Paper (Light)…

View on GitHub

GitHub Repository: https://github.com/GSuryaP/LazyDeskAI
License: MIT
Tech Stack: Python 3.10+, Streamlit, Google Gemma 2 (via Ollama), SQLite WAL, pypdf, pytest.

How I Built It

LazyDesk AI is engineered with modular, production-grade Python components built around open-source AI:

Local Inference Engine (core/llm_service.py): Connects directly to local Ollama running Google Gemma 2 (2B). It formats structured extraction prompts requesting JSON responses for executive summaries, to-dos, and date strings, with resilient heuristic fallbacks to ensure offline operation never hangs.
Document Extraction Pipeline (core/doc_intel.py): Extracts text from PDFs (pypdf), Markdown, text notes, and code, calculating word counts, line counts, and metadata before querying Gemma. Results are cryptographically cached in SQLite so re-reading a file is instantaneous.
Cryptographic Deduplicator (core/duplicates.py): Uses a two-stage filter: first clustering by exact byte size to eliminate 95% of unique files instantly, then computing full SHA-256 checksums on matching sizes to guarantee zero false positives.
Safe Organizer & Collision Engine (core/organizer.py): Proposes clean folder routing. If a target file already exists, it calculates non-destructive incremental suffixes (e.g., file_1.ext) and alerts the user.
Transactional Audit Log & Physical Undo (core/history.py & core/db.py): Backed by SQLite in WAL (Write-Ahead Logging) mode. The undo_operation function reverses moves on the actual file system, safely recreating parent folders if necessary.
Editorial Design System (ui/styles.py & ui/components.py): Custom theme-token system rendering a responsive Streamlit layout with persistent Light/Dark state across browser refreshes.

Why Does Open Innovation Matter?

The challenge prompt asks: Does your open-based approach work better than a closed one? What did it make possible that a closed API wouldn't?

For a personal desktop assistant, the answer is an emphatic yes. Building this tool with a proprietary cloud API (such as OpenAI or Anthropic) would be a fundamentally flawed product decision. Here is why open innovation is mandatory:

1. Absolute Air-Gapped Privacy

Your Desktop and Downloads folders contain your most private life: tax filings, bank statements, client NDAs, medical records, private keys, and personal photos.

Closed Cloud API: Uploading the full text of every downloaded PDF or contract to a third-party cloud server introduces immense security and compliance liabilities.
Open-Source Gemma 2: Google Gemma runs 100% on your machine. Zero tokens ever leave your computer. The application works flawlessly in airplane mode with no internet connection.

2. Zero Operating Cost

Closed APIs charge per thousand tokens. Running document analysis and OCR across hundreds of messy multi-page documents would quickly burn through expensive credits.
With open-weight Gemma running locally, organizing 10,000 files costs $0.00 forever.

3. True Model Sovereignty

Because LazyDesk AI interfaces with Ollama's open API standard, users have complete sovereignty. You can run lightweight gemma2:2b on a thin laptop, scale up to gemma2:9b or gemma2:27b on a workstation, or plug in a specialized fine-tune without changing a single line of application code.

The Lazy Payoff

How much effort does LazyDesk AI actually save?

Before: Spending 45 minutes every Friday clicking through a cluttered Downloads folder, manually dragging PDFs into folders, accidentally deleting duplicate invoices you might still need, and opening multi-page documents just to check if there was a deadline inside.
After: One click on "Scan Folder", one click on "Execute Moves", and you have a structured, tidy workspace. In seconds:
- 72 cluttered files sorted into categorized folders.
- Duplicate sets safely archived into _Duplicates_Archive/, reclaiming disk space.
- Important deadlines and to-dos extracted onto your dashboard by Gemma 2.
- Any accidental move reversible via the Undo Move button.

Do the smart work once, and let local open-source AI handle the rest.

Prize Categories

Primary Category: Best Use of Gemma ($200 USD) (LazyDesk AI uses Google's open-weight Gemma 2 model running locally on-device to power its document intelligence, executive summarization, action item extraction, and milestone date parsing with zero cloud latency and 100% air-gapped privacy).
Overall Challenge: Week 3 Winner ($250 USD)

Built with Google Gemma 2, Streamlit, and Python for the Hacktoberfest 2026 DEV Challenge.

DEV Community: G Surya Prakash