DEV Community: Mehul Patil

SlothSort — Let Your Files Organize Themselves

Mehul Patil — Fri, 23 Oct 2026 08:00:24 +0000

What I Built

SlothSort is a local-first AI file organization assistant that takes the boring work out of managing a cluttered Downloads folder.

Instead of manually opening files, figuring out what they contain, renaming them, and moving them into folders, SlothSort analyzes files and proposes meaningful names, categories, and destinations automatically.

It can group documents by topic and type, identify duplicate files, detect potential naming conflicts, and generate a cleanup plan for review before changes are applied. Its undo functionality makes it easy to reverse operations if something doesn't look right.

The goal is simple: spend less time managing files and more time doing the things that actually matter.

Whether you're a student buried under assignment PDFs, a developer drowning in downloads, or someone whose desktop has become a digital junk drawer, SlothSort helps you reclaim your time.

Demo

🎬 Demo video: [Add your demo video link]

🖥️ Live demo: [Add a deployed demo link, if available]

SlothSort is designed primarily as a local desktop application, so the most meaningful demo will show it organizing a real test folder on a local machine.

Code

💻 GitHub repository: [Add your GitHub repository link]

How I Built It

SlothSort is built around open-source AI and local inference.

Python: File inspection, organization logic, duplicate detection, and filesystem operations.

Ollama: Runs a compatible open-weight language model locally.

Open-weight LLM: Analyzes extracted file information and suggests descriptive filenames, categories, and destinations.

Local metadata and content extraction: Helps the model understand documents without relying on a cloud AI service.

Safety and reversibility: A preview-first workflow, conflict checks, and an operation history help keep file changes under user control.

The AI is used where semantic understanding matters, while deterministic code handles filesystem operations. This keeps the system more predictable than allowing a language model to directly manipulate files without safeguards.

The intended workflow is straightforward: scan, understand, suggest, review, organize, and undo when needed.

Why Does Open Innovation Matter?

Your files are personal. Your assignments, financial documents, work files, and private notes shouldn't need to be uploaded to a third-party server just to organize them.

Open-weight models make it possible to run SlothSort locally, giving users greater control over their data and how their AI assistant behaves.

Open innovation also makes the project adaptable. Users can experiment with different compatible models, inspect the organization logic, change categorization rules, and improve the tool for their own workflows instead of being locked into a single proprietary API.

Local inference can also eliminate per-request AI API charges, making repeated file organization practical without a recurring model-usage bill. After the model and required dependencies are installed, the core workflow can operate offline.

For me, the most important part of open-source AI is that it makes useful automation more accessible. You shouldn't need an expensive subscription or surrender control of your files to stop wasting time on repetitive digital housekeeping.

Less file management. More free time. Stay slow, stay organized. 🦥

MenuMind — A Local AI Translator for Restaurant Menus

Mehul Patil — Tue, 13 Oct 2026 11:58:04 +0000

Small restaurants often serve customers who speak different languages, but translating a menu isn't as simple as running it through a generic translator. Regional dish names can lose their meaning, descriptions can become awkward, and prices or important details must never change.

MenuMind is an AI-powered menu translation tool designed for local restaurants, cafés, and food businesses. It helps restaurant owners turn existing menus into multilingual, customer-friendly versions while preserving the original dishes, prices, and culinary context.

With MenuMind, restaurant owners can upload a menu, select a target language, review AI-generated translations, and export a translated menu ready to print or share digitally.

The goal is to make professional multilingual menus accessible to small businesses without requiring expensive translation services or technical expertise.

Who would pay for it?

Local restaurants and cafés serving multilingual customers.

Small food businesses that frequently update their menus.

Restaurant owners who want to reach tourists and customers from different linguistic backgrounds.

Instead of hiring a translator for every menu update, businesses could use MenuMind to produce an initial translation and review it before publishing.

Demo

Live Demo: [Add your deployed demo URL]

Demo Video: [Add your demonstration video URL]

The demo will showcase the menu upload, translation, review, and export workflow.

Code

GitHub Repository: [Add your GitHub repository URL]

The project is designed around open-source AI components, with a focus on local inference and customizable translation workflows.

How I Built It

MenuMind combines a lightweight web application with an AI-powered translation pipeline.

The planned architecture includes:

Frontend: React and Vite for menu uploads, language selection, translation previews, and editing.

Backend: Python and FastAPI to coordinate extraction, translation, validation, and export.

Menu extraction: OCR using an open-source engine such as PaddleOCR or Tesseract to extract text from menu images and PDFs.

AI translation: An open-weight multilingual language model running locally through Ollama or Transformers.

Culinary context: A customizable glossary for regional dishes, ingredients, cooking methods, and terms that should remain untranslated.

Validation: Python-based checks to compare item counts and prices against the original menu and flag uncertain translations.

Export: PDF generation for printable menus and digital sharing.

One important design decision is to separate translation from validation. A language model should not be trusted to preserve every number or food-related detail on its own. MenuMind can compare extracted prices and item lists against the original and flag discrepancies for human review.

The initial target is English-to-Marathi and English-to-Hindi translation, with support for additional languages as the project develops.

Why Does Open Innovation Matter?

For MenuMind, open innovation is not just about using a free library. It influences how the product can be built, deployed, and adapted to the needs of small businesses.

Local inference can protect customer data

Restaurants can process menus on their own computers rather than sending every document to a third-party AI API. This is particularly useful for businesses that prefer to keep their internal documents private.

Lower operating costs make the business model more practical

A hosted translation API may charge for usage. Running a suitable open-weight model locally can reduce recurring inference costs, making it easier to offer an affordable tool to small restaurants. Local processing still has hardware and electricity costs, but it avoids mandatory per-request model charges.

Models can be customized for regional cuisine

Generic translation systems may struggle with regional dishes, transliterated names, and culturally specific food descriptions. An open approach allows the translation pipeline to use custom glossaries, specialized prompts, and potentially fine-tuned models as suitable training data becomes available.

No dependence on a single AI provider

The translation component can be replaced with another compatible model as multilingual capabilities improve. The application does not need to be permanently tied to one provider's API, pricing, or availability.

The product can work without an internet connection

With the required model and dependencies installed, the core translation workflow can be designed to run offline. This is a capability I want to prioritize and verify in the finished application.

The larger idea is simple: small businesses should be able to benefit from AI without needing an expensive subscription or giving up control over how the technology works.

My First User

The first users I want to reach are local restaurant and café owners.

My plan is to take an existing menu, generate a translation, and ask the owner to compare it with their current menu. I want to learn whether the tool saves them time, which translations need correction, and whether they would pay for future menu updates.

I will use that feedback to prioritize improvements rather than building features based only on assumptions.

MenuMind is my attempt to turn open-source AI into something practical: a tool that helps small businesses communicate better with their customers while retaining control over their data and AI infrastructure.

NatureQuest — Turn a Walk Into an AI Adventure

Mehul Patil — Wed, 07 Oct 2026 13:39:15 +0000

What I Built

NatureQuest is an AI-powered outdoor scavenger hunt that turns an ordinary walk into a real-world exploration game.

Instead of giving people another reason to stay on their phones, NatureQuest gives them a reason to put the phone down and go outside.

Before heading out, users can generate a short nature quest such as:

🌳 Find a tree with unusual bark
🌼 Find a flower with five petals
🐦 Spot a bird with a long tail
🍂 Find three different types of leaves
🪨 Find an unusual rock

Once outside, the user only needs their camera when they want to verify a discovery. The AI identifies what they found and marks the challenge complete.

The screen is the shortest part of the experience — the real game happens outside.

NatureQuest is designed for students, families, hikers, casual walkers, and anyone who wants a simple reason to explore their surroundings.

demo link : naturequest4466.vercel.app

*ARCHITECTURE : *
🌿 NATUREQUEST
│
▼
┌───────────────────┐
│ Web / Mobile UI │
│ │
│ • Start Quest │
│ • Choose Time │
│ • Choose Difficulty│
│ • Camera │
└─────────┬─────────┘
│
┌─────────▼─────────┐
│ Quest Engine │
│ │
│ Activity + Time │
│ Difficulty + Area │
└─────────┬─────────┘
│
▼
┌────────────────────────┐
│ OPEN-WEIGHT LLM │
│ │
│ Generate Nature Quest │
│ + Tasks + Hints │
└───────────┬────────────┘
│
▼
┌─────────────────────────┐
│ QUEST UI │
│ │
│ 🌳 Find unusual tree │
│ 🐦 Find a bird │
│ 🍂 Find 3 leaves │
└───────────┬─────────────┘
│
👤 USER GOES OUTSIDE
│
▼
📷 CAMERA
│
▼
┌───────────────────────────┐
│ OPEN VISION MODEL │
│ │
│ • Object recognition │
│ • Plant / bird / nature │
│ • Quest verification │
└─────────────┬─────────────┘
│
Detection
│
▼
┌───────────────────────────┐
│ QUEST VALIDATOR │
│ │
│ Does discovery match │
│ the current challenge? │
└─────────────┬─────────────┘
│
┌───────┴────────┐
│ │
YES NO
│ │
▼ ▼
✓ Complete "Try again"
│
▼
┌───────────────┐
│ XP / Progress │
│ Badges / Streak│
└───────┬───────┘
│
▼
🌳 NEXT QUEST

How I Built It

NatureQuest is built around open-source AI, rather than using a closed AI API as the core of the experience.

The system has two main AI components:

Local/Open-Weight LLM

The language model generates personalized outdoor quests based on things such as:

available time
difficulty
environment
interests
type of activity

Open Vision Model

When the user discovers something outdoors, the vision model analyzes the camera image and determines whether it matches the quest.

A simplified flow is:

User chooses activity
↓
Open-weight LLM
↓
Generate Nature Quest
↓
🌳
User goes outside
↓
Finds something
↓
📷 Camera
↓
Open Vision Model
↓
Discovery verified ✓
↓
Next Quest

The project is designed so that the AI components can be run locally and swapped for other open models as they improve.

*Why Does Open Innovation Matter?
*
For NatureQuest, open AI isn't just a technical choice — it changes what the product can do.

A closed vision API would mean sending photographs of the user's surroundings to an external service. With an open model running locally, the recognition step can happen on the user's own device.

That provides:

🔒 Privacy — images don't have to leave the device
📡 Offline potential — the core experience can work without a constant internet connection
🔄 Model freedom — different open models can be swapped in and compared
🛠️ Customization — the vision model can eventually be fine-tuned for local plants, birds, and environments
💰 Low running cost — no per-image API charges for the core AI
🌍 Accessibility — the experience isn't dependent on having access to a proprietary AI service

Most importantly, open AI makes it possible to build an experience where AI helps people interact with the physical world without requiring the physical world to be uploaded to someone else's server.

That's what I think open innovation should enable: not another chatbot window, but technology that gets out of the way and lets people experience the world around them.