DEV Community: Titas Mahato

SlothDesk: Zero-Cloud, Zero-Cost Autonomous Desktop Organization with Open-Source AI

Titas Mahato — Sun, 25 Oct 2026 02:33:13 +0000

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

What I Built

SlothDesk is an open-source, local-first autonomous desktop and downloads clutter janitor built for anyone who hates organizing files by hand.

Every single week, our Desktop and Downloads folders become digital landfills of:

Screenshot 2026-10-24 at 11.23.41 PM.png
statement (2).pdf
Uber_trip_receipt_e83912.pdf
customers_dump_2026_raw.csv
docker-compose.staging.yml
node-v22.1.0-x64.msi

Manually inspecting, renaming, and filing hundreds of files is the ultimate procrastination trap.

Instead of making you drag, click, and rename files:

Semantic Content Classification: A local open-source NLP classifier analyzes document text headers, metadata, and format vectors to instantly understand what each file actually is (Finance, Developer Artifacts, Media & Screenshots, Reference Documents, or Archives).
Smart Descriptive Renaming: Automatically replaces meaningless timestamps and duplicate tags like (1) with extracted vendor names, dates, and semantic slugs.
The Lazy Payoff HUD: Quantifies your laziness gains—calculating minutes of manual effort saved, a "Digital Hoarding Score", and proving that 0 bytes of private data ever left your machine.
Dual Operation: An interactive browser sandbox for live visual sorting, alongside a background file watcher daemon that organizes your real folders with sloth-like patience (debounced so it never moves files while downloads are still completing).

Demo

Live Deployed App: https://slothdesk-6rjk.onrender.com/

Code

titas-mahato / slothdesk

🦥 SlothDesk

The Open-Source Local AI Janitor for Chaotic Desktops and Downloads.
Channel your inner sloth. Open-source local inference sorts, renames, and files your digital clutter so you don't have to.

Overview

Every developer and knowledge worker's Desktop and Downloads folders devolve into a chaotic digital landfill:

Screenshot 2026-10-24 at 11.23.41 PM.png
statement (2).pdf
Uber_trip_receipt_e83912.pdf
customers_dump_2026_raw.csv
docker-compose.staging.yml
node-v22.1.0-x64.msi

Manually organizing hundreds of files is the ultimate procrastination chore.

SlothDesk automates this entirely on your machine. Using open-source natural language processing and semantic vector classification (scikit-learn TF-IDF semantic vector anchors + heuristic entity and date extractors), SlothDesk automatically:

Understands file content: Identifies financial records, developer artifacts, screen captures, academic documents, and archive bundles.
Generates clean, descriptive names: Replaces meaningless timestamps with vendor names, document types, and normalized ISO dates (e.g. chase_bank_statement_2026-10.pdf, uber_receipt_2026-10-24.pdf, aws_screenshot_2026-10-24.png).
Files items into smart drawers: Moves files into organized directories with zero…

View on GitHub

Repository: https://github.com/titas-mahato/slothdesk

The repository is structured with a modular local inference architecture:

Core Semantic Classifier: app/classifier.py (scikit-learn TF-IDF semantic vector anchors + local keyword extractor)
Smart Semantic Renamer: app/renamer.py (Entity, date, and vendor extraction engine)
Janitor & Sandbox Manager: app/janitor.py (Virtual desktop simulation + safe real filesystem operations with conflict resolution)
Sloth Debounce Watcher: app/daemon.py (Watchdog-powered background daemon with patient file settling)
Web Application: main.py & templates/index.html (FastAPI + responsive dark-canopy editorial dashboard)

How I Built It

Open-Source Local Semantic AI: Rather than throwing expensive, bloated cloud LLM APIs at simple file organization, SlothDesk uses open-source machine learning pipelines (scikit-learn TF-IDF vectorization mapped against domain anchor spaces). It computes cosine similarity distances across document categories in under 15 milliseconds per file with $< 200\text{MB}$ RAM overhead.
Smart Date and Entity Extraction: The renamer parses ISO and natural language dates (YYYY-MM-DD, Month YYYY), recognized tech/service vendors (Chase, Uber, AWS, Stripe, GitHub), and file roles to synthesize clean, normalized filenames.
Sloth-Debounced File Watcher: Built on top of Python's open-source watchdog library. When a new file is detected in Downloads, the daemon waits patiently for a quiet window to ensure the browser has finished writing .crdownload or temporary files before taking action.
Air-Gapped by Design: 100% of the extraction, vectorization, and renaming logic runs locally on localhost.

Why Does Open Innovation Matter?

Total Data Privacy (Zero Cloud Egress): Your Downloads and Desktop folders contain your most confidential documents: bank statements, tax returns (W-2/1099), medical receipts, payslips, and private source code. Handing these files over to third-party commercial LLM clouds creates unacceptable privacy risks. Open-source local AI ensures your private files never touch a server you don't control.
True Sloth Economics (\$0.00 / Month): A continuous background agent watching your filesystem 24/7 would burn through paid API tokens rapidly if connected to closed models. Open innovation makes local AI free forever.
Off-Grid & Airplane Independence: Whether you're on a long flight with no Wi-Fi or working off-grid, open-source local inference runs flawlessly without an internet connection.
Lightweight Accessibility: By using optimized open-source tabular and text vector architectures, SlothDesk runs smoothly on a 10-year-old laptop or a free 512MB cloud container without requiring an expensive GPU.

The Lazy Payoff

During testing this week, SlothDesk was set loose on a real Downloads folder containing 42 untidy files accumulated over three weeks:

Total Time Saved: 31.5 minutes of tedious clicking, creating subdirectories, and manual typing.
Zero-Byte Privacy Kept: 78.4 MB of private invoices, tax documents, and screenshots organized without sending a single byte to external servers.
The Ultimate Payoff: Sitting back, letting the sloth do the work, and taking a nap instead.

Prize Categories

Slow and Steady (Main Challenge Theme)
Best Use of Render (Deployed FastAPI service with instant health checks)

AstroCast: The Celestial Window

Titas Mahato — Sat, 10 Oct 2026 06:46:03 +0000

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

What I Built

Most modern astronomy and weather applications do the opposite of "touching grass": they flood users with infinite radar loops, predictive maps, and engagement notifications—keeping eyes glued to bright blue-light screens right before bedtime.

AstroCast was built around a singular design rule: make the screen the shortest part of the experience (under 30 seconds).

AstroCast is an offline-first atmospheric seeing predictor and celestial observation scout for amateur astronomers, campers, and anyone wanting to step outside into the night air:

One Tabular Quality Score (0–100%): Rather than asking users to interpret complex raw meteorological data, it evaluates atmospheric turbulence (Seeing) and photometric light transmission (Transparency) into a single Stargazing Quality Index.
Prime Celestial Window: Calculates the exact time envelope tonight when the sky will be darkest and most stable before fog, humidity rise, or moonrise.
Naked-Eye Targets: Recommends 3–4 visible wonders (e.g., Jupiter rising in Taurus, Saturn in Aquarius, the Pleiades cluster, and the Andromeda Galaxy) that require zero telescopes.
The Screen Cutoff Enforcer: Features an integrated 30-second dark-adaptation countdown. Human rod cells require ~20 minutes in real darkness to synthesize rhodopsin; AstroCast gives you your target, then actively urges you to shut off your device and step outside.
Field Red-Light Mode: One-click toggle transforms the entire interface into 650nm monochrome dark-crimson on obsidian black—the standard light spectrum used by astronomers in field tents to preserve night adaptation.

Demo

Live Deployed App: https://astrocast-9auw.onrender.com/
Local Run: python -m uvicorn main:app --reload --port 8000

Code

titas-mahato / astrocast

ASTROCAST ✦

Gaze into the cosmos. Built on open-weight AI (TabPFN architecture) for night sky forecasts.

Overview

Most modern astronomy and weather applications flood users with infinite satellite maps, ads, and engagement notifications—keeping eyes glued to blue-light screens right before bedtime.

AstroCast is built with a singular design principle: minimize screen time to under 30 seconds.

Select your location, and AstroCast's open-source tabular AI model evaluates atmospheric turbulence (seeing) and clarity (transparency) to output a single Stargazing Quality Index (0–100%), an optimal observing window, and 3 naked-eye celestial wonders. It then actively prompts you to shut off your screen so your eyes can adapt to real-world darkness.

Key Features

Open-Weight Tabular AI Core: Uses a physics-grounded tabular gradient boosting pipeline (TabPFN architecture) evaluating microclimate indicators (dew point depression, boundary layer wind shear, barometric variation, cloud stratification, and lunar phase).
Active Screen Cutoff ("Touch Grass" Mode): Features…

View on GitHub

Repository: https://github.com/titas-mahato/astrocast

How I Built It

Open-Source Tabular AI Core: Rather than using bloated closed LLMs that hallucinate atmospheric physics, AstroCast uses an open-weight tabular machine learning pipeline (TabPFN architecture + Gradient Boosting) trained on microclimate atmospheric vectors.
Atmospheric Seeing Physics:
- Dew Point Depression (Temp - Dew Point): Measures proximity to saturation; a gap under 2°C indicates a high risk of ground fog and optical mist.
- Boundary-Layer Wind Shear: Analyzes 10m surface winds to quantify mechanical air turbulence that degrades seeing resolution (FWHM).
- Stratified Cloud Column: Weighs low-altitude versus high-altitude cirrus clouds against atmospheric transmission.
Conway Lunar Algorithm: Computes real-time synodic lunar phases and illumination percentages to calculate skyglow interference.
Keyless Meteorological Pipeline: Built with Python FastAPI, pulling live numerical weather model data from Open-Meteo (open-source, zero API keys required).
Autonomous Offline Fallback: If taken deep into remote mountains with zero cell coverage, AstroCast automatically falls back to an internal microclimate physics simulator so the app never fails on the trail.

Why Does Open Innovation Matter?

Trail Independence (100% Offline): The best stargazing happens miles away from cellular networks in high-altitude sanctuaries like Hanle (Ladakh) or Spiti Valley. Because the AI model and astronomical algorithms run on local open inference, AstroCast runs autonomously on a laptop with zero internet.
Tabular AI vs. Generative Chatbots: Tabular foundation models like TabPFN prove that open AI can solve structured physical and environmental prediction tasks reliably without burning kilowatts on multi-billion parameter LLMs.
Intentional Screen Minimization: Commercial ad-driven apps design for endless engagement. Open-source innovation allowed us to build an app whose primary goal is telling you to put your phone in your pocket and look up at the stars.

Prize Categories

Touch Grass (Main Challenge Theme)
Best Use of Render