DEV Community: Rohit Kumar singh

slothos (week 3 DEv)

Rohit Kumar singh — Sat, 24 Oct 2026 15:33:37 +0000

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

What I Built

I built SlothOS, a local-first AI-powered file organization assistant designed to reduce digital clutter and save time on repetitive file management.

Instead of manually sorting files, searching through folders, and organizing documents one by one, SlothOS helps scan files, categorize them, identify duplicates, suggest better filenames, and organize files into relevant folders. It also includes an undo feature so users can reverse organization actions.

The goal is simple: let AI handle the boring digital housekeeping while you focus on what actually matters.

Demo

Demo Link: https://slothos-yer9.onrender.com/

Run the project locally to explore the dashboard and AI-powered file organization features.

Code

Source Code: https://github.com/rohit060321/slothos

The repository contains the React frontend, FastAPI backend, and integration with a locally running Ollama model.

How I Built It

SlothOS is built using React, Vite, Python, FastAPI, Ollama, and Gemma 3 1B.

React + Vite: Provides the interactive dashboard.
FastAPI: Handles backend APIs and file-management operations.
Ollama + Gemma 3 1B: Enables local AI inference without depending on a paid, cloud-hosted AI API.
File scanning and categorization: Helps organize files based on their content and characteristics.
Duplicate detection: Identifies files with matching content.
Filename suggestions: Helps make file collections easier to understand.
Undo functionality: Allows users to reverse supported organization actions.

The project follows a local-first approach, keeping AI inference on the user's machine when using the local Ollama setup.

Why Does Open Innovation Matter?

Open innovation made it possible to build an AI-powered productivity tool without relying entirely on expensive, closed AI APIs.

Using open-weight models through Ollama gives developers more control over inference, experimentation, and local execution. It also makes AI experimentation more accessible to students and independent developers who want to build useful tools without needing a large cloud budget.

SlothOS demonstrates how open-source AI can be applied to an everyday problem: reducing the time and effort spent managing digital files.

My Agent Session

I used AI-assisted development to plan, implement, and debug parts of SlothOS, including the React interface, FastAPI integration, and local model workflow.

GitHub Repository

Add your DevRelay agent session link here if you decide to share one.

The Lazy Payoff

SlothOS is built around a simple idea: your computer should spend less time making you manage your files.

By helping with categorization, duplicate detection, filename suggestions, and folder organization, it aims to reduce repetitive manual work and make cluttered directories easier to manage.

The bigger goal is to make digital organization feel less like a chore and more like something that quietly takes care of itself.

Prize Categories

Open-Source AI Project
Best use of open-weight AI / local inference, if eligible under the challenge's official category rules

Ghost Writer

Rohit Kumar singh — Sat, 17 Oct 2026 18:47:31 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle

What I Built

I built GhostWriter AI, an AI-powered sales content generator designed for students, freelancers, small business owners, and online sellers who want to grow their side hustles.

Many people have great products but struggle to create attractive marketing content. Hiring professional copywriters can be expensive, especially for beginners and small businesses.

GhostWriter AI solves this problem by generating a complete sales content kit from simple product details. Users enter their product name, category, features, target customers, price, offer, preferred platform, tone, and language.

The application generates:

Catchy product titles
Short and detailed product descriptions
Social media captions
Relevant hashtags
Promotional messages
Marketing hooks to attract potential customers

GhostWriter AI helps users save time, reduce content creation costs, and focus more on growing their businesses.

The project could be monetized through a freemium model, offering basic content generation for free and premium features through a paid subscription.

Demo

Try the live application here:

https://ghostwriter-czm9.onrender.com/

Users can enter their product details, generate AI-powered marketing content, and download the generated sales content as a Markdown file.

Code

GitHub Repository:

https://github.com/rohit060321/ghostwriter.git

The repository contains the source code, setup instructions, and documentation needed to run the project.

Contributions and suggestions for improving the project are welcome.

How I Built It

I built GhostWriter AI using Python, Streamlit, FastAPI, Ollama, and the open-weight Gemma 3 1B model.

Technologies used:

Python: Implements the application's core logic and input processing.
Streamlit: Provides a simple and interactive user interface.
FastAPI: Handles backend requests and communicates with the AI model.
Ollama: Runs the language model locally during development.
Gemma 3 1B: Generates product titles, descriptions, social media captions, hashtags, promotional messages, and marketing hooks.
Render: Hosts the deployed application.

How it works:

The user enters the product details and selects their preferred marketing settings.
The Streamlit frontend sends the information to the FastAPI backend.
The backend prepares a prompt based on the user's requirements.
The Gemma model processes the prompt and generates relevant marketing content.
The generated content is displayed to the user and can be downloaded.

I wanted to keep the project simple, practical, and accessible rather than building a complicated system with unnecessary dependencies.

Why Does Open Innovation Matter?

Open innovation makes AI development more accessible to students, independent developers, startups, and small businesses.

By using an open-weight model such as Gemma, developers can experiment with AI, customize prompts, understand how model-powered applications work, and build useful tools without depending entirely on proprietary AI APIs.

For GhostWriter AI, open-weight AI provides an opportunity to explore more accessible content generation, greater control over model execution, and the possibility of adapting the application to different business requirements.

Open-source technologies such as Python, FastAPI, Streamlit, and Ollama also make it easier for other developers to study, modify, and contribute to the project.

As a student developer, building this project helped me understand how an AI model can be integrated into a real application to solve a practical problem. It also showed me how open-source AI can help turn an idea into a useful tool for people trying to build their own side hustles.

My Agent Session

This project was developed with AI-assisted coding and experimentation. I focused on integrating the model, connecting the frontend and backend, and making the application usable.

Agent session link: Not available.

Prize Categories

Open-Source AI Project: Uses the open-weight Gemma model for AI-powered marketing content generation.

NPC irl

Rohit Kumar singh — Sun, 11 Oct 2026 15:11:17 +0000

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

What I Built

NPC IRL — Your world is the map. Your AI gives you side quests.**

What if your boring neighborhood was actually an open-world RPG waiting to be explored?

NPC IRL is a real-world adventure game powered by locally running AI. Instead of spending more time staring at a screen, you use AI to generate short, weird, and fun missions that get you outside and interacting with the world around you.

Examples of quests:
-Detective Mode: Find something in your neighborhood that looks suspiciously out of place.

Nature Explorer: Find three different leaf shapes.
Alien Mode: Find an everyday object an alien would completely misunderstand.
Chaos Mode: Find an object that looks like it knows a secret. Give it a name.

Complete quests, earn XP, level up, and turn an ordinary walk into a real-life adventure.

The goal is simple: make the screen the starting point of the adventure, not the destination.

Demo

https://npc-irl-dnra.onrender.com/

Code

https://github.com/rohit060321/npc-irl

The project uses a lightweight Python backend and a simple web frontend, making it easy to run and experiment with.

How I Built It

I built NPC IRL using a minimal stack to keep the project accessible and easy to modify.

Tech stack:

Python: Backend server and AI integration.
HTML, CSS, JavaScript: Interactive RPG-style interface.
Ollama: Local inference runtime.
Llama 3.2 3B: Open-weight language model for generating quests.
LocalStorage: Saves XP and completed quest counts in the browser.

The backend sends a prompt to the locally running model, which generates a quest containing a title, mission, XP reward, and fun fact.

The frontend displays the mission and tracks progress as the player completes quests.

The architecture is intentionally simple: no paid AI API, no complicated agent framework, and no cloud database required.

Once the model has been downloaded, quest generation can work without an internet connection, provided the local Ollama service is running.

Why Does Open Innovation Matter?

For a project like NPC IRL, open innovation isn't just a technical choice. It's part of the experience.

1. AI without a cloud dependency

A locally running model means quests don't need to be generated by sending every request to a remote AI provider. After downloading the model, the core quest-generation experience can work without internet access.

2. Privacy by design

Your quest prompts can stay on your own machine instead of being sent to a third-party AI API. The application doesn't need a cloud database to track basic game progress.

3. Freedom to experiment

I can change the system prompt, experiment with different quest personalities, or swap in another compatible local model without rebuilding the entire application around a proprietary API.

Want a chaotic game master? A nature expert? A mysterious detective? The model and prompts can evolve with the project.

4. Lower running costs

There are no per-request AI API charges. The trade-off is that local inference uses your machine's processing power and memory.

For an application designed to encourage offline exploration, this is a meaningful advantage.

The bigger idea: Open AI makes it possible to build small, personal experiences that don't need to depend on a centralized service. Technology can help us reconnect with the physical world without demanding more of our attention.

My Agent Session

I haven't published a DevRelay session yet.

The project can be explored through the source code in the GitHub repository.

Prize Categories

Open-source AI / open-weight models / local inference.

NPC -IRL

Rohit Kumar singh — Sun, 11 Oct 2026 15:01:45 +0000

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

What I Built

Demo

Code

How I Built It

Why Does Open Innovation Matter?

My Agent Session

Prize Categories

BUILD FOR FRIEND CHALLENGE

Rohit Kumar singh — Mon, 05 Oct 2026 06:38:21 +0000

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend

What I Built

I built SpeakBuddy, a private AI communication practice partner for a friend who gets nervous and loses confidence while speaking English during interviews, presentations, and conversations.

My friend often knows what they want to say but struggles to express their thoughts clearly and confidently. SpeakBuddy provides a safe space to practice without the pressure of a real interview or presentation.

The user can choose between Interview, Presentation, and Conversation modes. SpeakBuddy asks a question, analyzes the user's answer using AI, provides scores for Clarity, Grammar, Vocabulary, and Confidence, gives improvement suggestions, and generates a follow-up question for continued practice.

Demo

Live Demo: https://speakbuddy-dnra.onrender.com/

Video Demo: https://drive.google.com/file/d/1e8jhW-B7tqM9UsAffvUzhhPyfrWRvC8Z/view?usp=sharing

Code

GitHub Repository: https://github.com/rohit060321/SpeakBuddy.git

How I Built It

The core of SpeakBuddy is built around Gemma 3 1B, an open-weight AI model running locally through Ollama.

The application is built using Python and FastAPI for the backend, HTML/CSS/JavaScript for the frontend, MongoDB Atlas for storing practice sessions, ElevenLabs for voice feedback, and Render for deployment.

The basic flow is:

Question → User Answer → Gemma Analysis → Scores → Suggestions → Follow-up Question

Gemma analyzes the user's response and returns structured feedback containing scores for clarity, grammar, vocabulary, and confidence, along with improvement suggestions and a follow-up question.

Why Does Open Innovation Matter?

Open innovation was important for SpeakBuddy because communication practice can involve personal answers, experiences, and weaknesses.

Using an open-weight model like Gemma gave me more control over how the AI works instead of depending completely on a closed AI API.

Running Gemma locally through Ollama also made experimentation easier. I could modify prompts, change the feedback structure, test different behaviors, and build the application around the model rather than treating the AI as a black box.

For this project, open innovation made it possible to build an AI tool that is more customizable, transparent, and privacy-conscious.

My Agent Session

I used AI-assisted development while building SpeakBuddy to help with implementation, debugging, and improving the project.

Prize Categories

Gemma / Hugging Face — Gemma 3 1B is used as the core AI model through Ollama.
MongoDB Atlas — used for storing practice sessions and history.
ElevenLabs — used for voice feedback.
Render — used to deploy the application.

hellloooo everone

Rohit Kumar singh — Mon, 05 Oct 2026 01:29:50 +0000