DEV Community: Devansh Gupta

The Last Lantern – An Offline-First AI Storytelling Experience

Devansh Gupta — Fri, 30 Oct 2026 16:37:51 +0000

The Last Lantern: Stories, Memories & Open AI

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine

What I Built

I built The Last Lantern, an AI-powered storytelling experience that explores the creative side of artificial intelligence.

The idea was to build something different from a regular AI chatbot. I wanted to create a small space where technology could be used to tell stories, preserve meaningful memories, and explore folklore.

The application has three storytelling modes:

Haunted Archives: Generates mysterious stories inspired by supernatural themes, forgotten places, and unexplained events.
Memory Lantern: Turns personal memories and meaningful experiences into thoughtful written tributes.
Forgotten Legends: Creates folklore-inspired stories based on myths, traditions, and legendary places.

The project uses a locally running open-weight AI model instead of depending on a paid cloud AI API.

Demo

Live Website:

https://devanshguptacu-create.github.io/The-Last-Lantern/

The frontend is hosted using GitHub Pages. To generate stories, users need to run Ollama and the required AI model locally on their computer.

Code

GitHub Repository:

https://github.com/devanshguptacu-create/The-Last-Lantern

The project currently uses three main files:

index.html – Application structure
style.css – Visual design and responsive layout
script.js – Application logic and local AI integration

How I Built It

I built the frontend using HTML, CSS, and JavaScript. For the AI part, I chose Ollama with the Qwen 2.5 3B open-weight model.

The application communicates with Ollama's local API. When a user selects a storytelling mode and enters a prompt, JavaScript prepares the request and sends it to the locally running model. The generated story is then displayed on the website.

I also added browser LocalStorage to save generated stories, along with options to copy and download them.

My main technology choices were:

HTML5 and CSS3 for the interface
JavaScript for application functionality
Ollama for running AI locally
Qwen 2.5 3B for story generation
LocalStorage for saving stories in the browser

One interesting part of building this project was connecting a static frontend to a locally running AI model without creating a separate backend server.

Why Does Open Innovation Matter?

For me, open innovation means having the freedom to experiment, learn, and build without being limited to a single closed platform.

I chose Ollama and an open-weight model because I wanted to understand how local AI actually works, rather than simply connecting an application to a paid API.

This approach makes a few things possible:

Developers can experiment with AI models locally.
Users do not need a paid cloud API subscription for story generation.
Developers can explore different open-weight models.
Prompts can be processed locally instead of being sent to a third-party AI service.

Building The Last Lantern helped me understand that open AI is not only useful for research or complex technical applications. It can also be used to create simple, creative experiences.

The Spirit of It

The name The Last Lantern represents the idea that a story or memory can continue to have meaning even when time passes.

I wanted the project to feel a little mysterious while also having a meaningful side. The Haunted Archives and Forgotten Legends explore imagination, while Memory Lantern focuses on preserving personal experiences through writing.

This project is my attempt to combine storytelling and local AI into one experience.

Developer

Devansh Gupta

B.Tech CSE | AI & ML Enthusiast

GitHub: https://github.com/devanshguptacu-create

Built with curiosity, creativity, and open AI. 🕯️

SlothSort AI Let AI Do the Sorting. You Do the Chilling.

Devansh Gupta — Tue, 20 Oct 2026 10:14:55 +0000

SlothSort AI: Let AI Do the Sorting. You Do the Chilling.

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

What I Built

Meet SlothSort AI — a simple, AI-powered file organizer designed for people who want to spend less time organizing files and more time doing what matters.

We all know the feeling: a Downloads folder full of documents, images, videos, code files, and random files we promise to organize later. Manually sorting everything is repetitive, boring, and easy to procrastinate on.

SlothSort AI makes that process easier. Choose a folder, let the app analyze filenames, review the suggested categories, and organize the files into folders such as Documents, Images, Videos, Audio, Code, Archives, and Other.

The goal is simple: less manual work, less digital clutter, and more time for the things you actually enjoy.

Who gets to be a little lazier? Students, developers, creators, and anyone whose folders have become a digital mess.

Demo

Try SlothSort AI live:

https://devanshguptacu-create.github.io/SlothSort-AI/

The project is designed to run in a compatible desktop browser. Folder access requires browser support and your permission. AI-powered classification also depends on a working Hugging Face Inference connection.

Code

GitHub Repository:

https://github.com/devanshguptacu-create/SlothSort-AI

The repository contains the HTML, CSS, and JavaScript used to build the project.

If you explore the code, you can see how the interface, file handling, AI classification, and category-based organization work together.

How I Built It

I built SlothSort AI using HTML, CSS, and JavaScript, keeping the project lightweight and accessible without requiring a separate Python backend or a local model installation.

The main pieces are:

HTML: Structures the dashboard and file-organizing interface.
CSS: Creates a modern, sloth-inspired interface with a dark visual theme.
JavaScript: Handles folder selection, file scanning, classification, and organization.
Open-weight AI model: The project uses Hugging Face Inference with facebook/bart-large-mnli for zero-shot text classification of filenames into relevant categories.
Browser File System Access API: Allows supported browsers to access a folder after the user grants permission.

The AI connection is an important part of the project: instead of relying only on fixed file extensions, the app can use a language model's classification capabilities to suggest categories from filenames. Extension-based categorization can serve as a fallback when AI inference is unavailable.

The hosted inference approach keeps the project easy to try without installing a large model locally. It does, however, depend on network access, provider availability, and a valid Hugging Face access token where required.

Why Does Open Innovation Matter?

Open innovation makes experimentation more accessible.

By building around an open-weight model, I could explore AI-powered classification without training a model from scratch or building an entire inference infrastructure myself. Hugging Face also makes it easier to experiment with models and connect them to practical projects.

For a student developer, this matters. It lowers the barrier to learning, prototyping, and building useful tools with AI.

It also gives developers more flexibility to inspect the model's documentation, experiment with alternatives, and adapt the implementation as their needs grow.

SlothSort AI represents my belief that AI should not only solve complicated problems. It should also eliminate small, repetitive frustrations from everyday life.

Open innovation helps make that kind of experimentation possible for more people.

My Agent Session

I haven't included an agent session link in this submission. This section is optional, so I'll add one if I have a shareable DevRelay session available.

The Lazy Payoff

The intended payoff is straightforward: fewer manual sorting decisions and less repetitive file management.

Instead of creating folders and moving every file individually, users can review AI-suggested categories and organize files in batches.

I haven't measured a verified time saving yet, so I won't invent a number. The next step is to test SlothSort AI on a real, messy folder and compare the time taken with manual organization.

The philosophy: automate the boring parts, keep the human in control, and save your energy for something better.

Always review the suggested categories before moving important files, and test with copies of files first.

Prize Categories

My primary focus is the Slow and Steady challenge: using open-source AI to reduce repetitive effort and make everyday work easier.

I'll select any additional partner prize categories only if the project meets their published eligibility requirements.

Built with curiosity, powered by open innovation, and inspired by the art of doing less repetitive work.

If you like the idea, check out the repository, try the demo, and share your feedback!

Hacktoberfest #OpenSource #AI #JavaScript #BuildInPublic

SideHustle Studio: Create Better Product Listings with Local Open-Weight AI

Devansh Gupta — Thu, 15 Oct 2026 06:16:19 +0000

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

What I Built

SideHustle Studio — Your product deserves a better story.

SideHustle Studio is an AI-powered product listing writer designed for small business owners, freelancers, online sellers, and side hustlers.

Many small sellers spend too much time writing attractive product descriptions for their online shops and social media pages. Professional copywriting can be expensive, and hosted AI tools may introduce recurring costs.

SideHustle Studio aims to make this process simpler by generating an editable first draft from seller-provided product details.

Key Features

Generate product titles and descriptions.
Prepare listing drafts for platforms such as Instagram, Etsy, eBay, and Shopify.
Choose English, Hindi, or Hinglish output.
Select a writing tone and target audience.
Include a seller-provided price.
Copy or download the generated draft.
Use a local AI model rather than relying on a paid hosted AI API.

The goal is to help small sellers spend less time writing listings and more time growing their businesses.

Demo

Live Website: SideHustle Studio

GitHub Repository: View the source code

Deployment note: GitHub Pages hosts the static frontend. The local AI generation feature requires the Python backend and Ollama model to be running, so it may not work on the public GitHub Pages website.

Code

The project uses HTML, CSS, and JavaScript for the frontend, with Python and Flask connecting the application to a local Ollama model.

Repository: https://github.com/devanshguptacu-create/SideHustleStudio.git

How I Built It

I built SideHustle Studio around a practical problem: helping independent sellers create useful product listing drafts without depending on a paid hosted AI API.

The project uses:

HTML5 for the page structure.
CSS3 for a responsive, nature-inspired green-and-cream interface.
JavaScript for form interactions, loading states, copying, and downloading drafts.
Python and Flask for the backend.
Ollama to run the language model locally.
Qwen 2.5 3B as the default model for local inference.

The backend sends seller-provided product information to the locally running model and returns a draft containing a title, description, key benefits, call to action, and tags.

The prompt also instructs the model not to invent product specifications, certifications, guarantees, or other unsupported claims.

The model needs to be downloaded, and the backend must be running for AI generation to work. The public GitHub Pages deployment currently serves the frontend only.

Why Does Open Innovation Matter?

Open innovation gives developers the freedom to experiment, inspect prompts, adapt workflows, and explore different models without depending entirely on a closed AI provider.

For a small-business tool, local inference can reduce dependence on per-request API charges and allow product information to be processed on the user's own machine.

It also gives developers more control over how the system behaves and makes it easier to customize the tool for a particular business niche.

Local AI is not automatically free of every cost: users still need suitable hardware, storage, and electricity, and model licenses must be respected.

For me, the value of open-source AI is the opportunity to build useful tools that are more customizable and potentially more accessible to people starting small businesses.

My Agent Session

No public agent session is included in this submission yet.

My First User

The next step is to share the prototype with a small seller, observe how they use it, and collect honest feedback about listing quality, time saved, and the features they would find valuable.

I will update this section when I have real user feedback.

Prize Categories

Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle.

Final Thoughts

I believe AI should make it easier for people to turn their skills and ideas into real opportunities.

SideHustle Studio is a step toward making practical AI tools more accessible to small sellers, without making a paid hosted AI service a requirement for every generation.

SideHustle Studio — Small business energy. Open AI tools. ✨

Built by Devansh Gupta.

TouchGrass AI: Less Scrolling, More Living — Turning Screen Time into Real-World Adventures

Devansh Gupta — Fri, 09 Oct 2026 11:04:15 +0000

🌿 TouchGrass AI — Less Scrolling, More Living

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

What I Built

TouchGrass AI is a nature-inspired web application designed to help people spend less time staring at screens and more time exploring the real world.

We spend hours scrolling through our phones, often forgetting the simple experiences waiting outside. TouchGrass AI aims to change that by turning the idea of going outdoors into small, enjoyable, and achievable adventures.

🌱 What can you do with TouchGrass AI?

🌳 Personalized outdoor missions: Discover activities based on your mood, available time, and interests.
🎯 Simple challenges: Turn free time into an opportunity to explore nature.
📚 Adventure history: Keep track of completed outdoor activities.
❤️ Save your favorites: Bookmark missions you want to try later.
📱 Responsive design: Enjoy a clean, nature-inspired interface across different screen sizes.
🔒 Browser-based storage: Save your missions and activity history locally using localStorage.

The project is designed for students, busy professionals, nature lovers, and anyone who wants to build healthier digital habits.

The main idea is simple: use technology as a bridge to the real world, not as a replacement for it.

The current version is a functional front-end prototype. Its mission recommendations use predefined JavaScript rules; integration with an actual open-weight AI model is a planned next step.

Demo

🌐 Live Website: https://devanshguptacu-create.github.io/TouchGrass-AI/

Explore the website, select your preferences, discover an outdoor mission, and take your first step toward spending more time outside!

Code

💻 GitHub Repository: https://github.com/devanshguptacu-create/TouchGrass-AI

The project is built with standard web technologies:

HTML5 — Application structure
CSS3 — Responsive design and nature-inspired styling
Vanilla JavaScript — Mission generation, navigation, and user interactions
localStorage — Local persistence for saved missions and activity history

To run the project locally:

git clone https://github.com/devanshguptacu-create/TouchGrass-AI.git
cd TouchGrass-AI

Then open index.html in your browser.

How I Built It

I built TouchGrass AI with a simple principle in mind: the screen should be the starting point of the experience, not the destination.

I used HTML, CSS, and vanilla JavaScript to create an interactive, responsive interface. Users can choose their preferences, discover suitable outdoor activities, save missions, and track their progress.

The current recommendation system uses a predefined collection of activities and JavaScript logic to suggest missions. Browser localStorage keeps saved activities and history available between visits on the same browser.

What about open-source AI?

The current version does not yet integrate an open-weight AI model or local inference framework. Adding an open-source model for more flexible, context-aware outdoor recommendations is a planned development goal.

Potential next steps include exploring open-weight models and browser-based inference tools such as Transformers.js, subject to model compatibility, performance, and licensing requirements.

I want to develop the project toward a system where AI helps people discover meaningful real-world experiences without making screen time the main activity.

Why Does Open Innovation Matter?

Open innovation makes it possible for developers to experiment, learn, collaborate, and build on shared technology rather than starting from scratch.

For a project like TouchGrass AI, open-source tools could eventually make it possible to explore local AI inference, customize recommendation behavior, and improve privacy without relying entirely on a closed, proprietary AI API.

It also makes the project easier for other developers to inspect, improve, and extend with ideas such as offline support, nature exploration challenges, garden planning, or bird-call identification.

I believe useful technology should not only be powerful; it should also be accessible, understandable, and open to improvement.

My goal is to build technology that encourages people to experience more of the world beyond their screens.

My Agent Session

Agent session: Not included in this submission yet.

I plan to document future AI-assisted development sessions as the project evolves toward open-weight AI integration.

Prize Categories

This project is submitted for Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass.

Additional partner prize categories may apply once the project meets their specific requirements. The current version does not claim integration with any particular AI partner or framework.

🌍 Final Thought

We build technology to improve our lives. Sometimes, that means building something that encourages us to put our phones down.

TouchGrass AI — Discover your next adventure. Then go live it. 🌿

Built by Devansh Gupta.