DEV Community: abhimanyu sharma

Ghostlight AI — Stories from the Other Side

abhimanyu sharma — Sat, 31 Oct 2026 17:44:23 +0000

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

What I Built

👻 Ghostlight AI — Stories from the Other Side

For the Ghost in the Machine challenge, I built Ghostlight AI, a spooky, browser-based AI storytelling experience inspired by Halloween, Día de los Muertos, and the stories and memories that stay with us.

The project has two main features:

🎃 Ghost Story Generator: Create mysterious, spooky stories based on a selected mood and location.
🕯️ Memory Keeper: Write meaningful remembrance messages to celebrate loved ones and preserve special memories.

I wanted to combine the playful spirit of Halloween with something more meaningful: using technology not only to create fictional stories but also to help people express and preserve memories.

The interface features a magical ghost, a haunted atmosphere, glowing purple visuals, candles, marigolds, and Día de los Muertos-inspired artwork.

My goal was to make AI storytelling accessible through a simple website without requiring users to install a local AI runtime.

Demo

🌐 Live Project: Ghostlight AI

🖼️ Project Poster: The poster showcases the Ghostlight AI experience, its spooky storytelling interface, and the Memory Keeper feature.

Explore the project and experience a little magic from the other side!

Code

💻 GitHub Repository: Ghostlight AI on GitHub

The project is built using HTML, CSS, and JavaScript, with an open-source AI model integrated into the browser.

Feel free to explore the code, suggest improvements, and contribute to the project!

How I Built It

I kept the architecture simple and focused on accessibility.

Technology stack:

HTML5: Structures the storytelling interface.
CSS3: Creates the Halloween-inspired design, glowing effects, and responsive layout.
Vanilla JavaScript: Handles user interactions, story generation, and the Memory Keeper experience.
Transformers.js: Provides browser-based inference using a compatible open-source model.
FLAN-T5 Small: The open-weight language model used for text generation.
Hugging Face model ecosystem: Provides access to the model used by the application.

The model is loaded in the browser through Transformers.js, using the Xenova/flan-t5-small model.

The application does not require a paid AI API key or a custom application server for inference. The model files must be downloaded when first loading the model, so an internet connection is required initially. Actual offline availability depends on whether the required assets are cached and accessible.

I also included a fallback storytelling mechanism so users can still explore the interface if the model fails to load. The fallback uses predefined generation logic rather than AI inference.

The result is a lightweight project that brings open-source AI into a fun, themed web experience.

Why Does Open Innovation Matter?

Open innovation matters because AI should be something developers can explore, understand, adapt, and build with—not just consume through closed services.

By using open-source tools and an open-weight model, I could experiment with browser-based inference and build an interactive experience without depending on a paid proprietary AI API.

This approach offers several benefits:

Accessibility: Developers can experiment without needing a paid AI API subscription.
Transparency: The project can be inspected, modified, and extended by the community.
Privacy potential: Browser-based inference can reduce the need to send prompts to a remote inference service, although privacy also depends on the application's implementation and its external dependencies.
Learning: Developers can explore how AI models integrate with ordinary web technologies.
Community collaboration: Contributors can improve the interface, generation logic, accessibility, and model integration.

Open innovation makes it easier to turn a creative idea into an experiment that other developers can learn from and build upon.

My Agent Session

I built this project by combining web development with open-source AI integration, focusing on making the experience simple and accessible.

I have not attached a DevRelay agent session to this submission. If I record one, I can add the session link here.

The Spirit of It

What if a ghost could tell a story—and help us remember the people who made our lives special?

That question inspired Ghostlight AI.

Halloween is often associated with spooky stories, haunted houses, and mysterious characters. Día de los Muertos adds another dimension: remembering and celebrating loved ones.

I wanted my project to bring these ideas together.

The Ghost Story Generator creates space for imagination, while the Memory Keeper offers a place to express remembrance. One celebrates fictional worlds; the other honors real memories.

For me, the spirit of Ghost in the Machine is about giving technology a human touch. AI can help us create something entertaining, but it can also help us express feelings, explore ideas, and keep meaningful stories alive.

Sometimes, the most memorable technology is the kind that helps us reconnect with our own humanity.

Prize Categories

My primary focus is the Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine.

If additional partner prize categories are listed on the official challenge page, I will select only those that match the project's actual technologies and implementation.

Thanks for checking out Ghostlight AI! 👻

If you like the project, visit the repository, share your feedback, and consider contributing.

Create Tales. Preserve Memories. Keep the Magic Alive. 🕯️🎃

LazyDesk AI: Do Less, Feel Caught Up with Open-Weight AI

abhimanyu sharma — Wed, 21 Oct 2026 18:00:38 +0000

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

What I Built

I built LazyDesk AI, a local-first AI productivity assistant designed to make everyday tasks feel less overwhelming.

We all have those days when our minds are full of unfinished work: replying to messages, completing assignments, organizing files, and remembering small personal tasks. Sometimes, even deciding where to start takes too much energy.

LazyDesk AI helps users unload their thoughts into a simple text box and transforms that messy list into a gentle, manageable plan.

Key features

🧠 Brain Dump to Tiny Plan: Turn scattered thoughts into a prioritized list with small, actionable steps.
✍️ AI Message Drafter: Turn rough notes into short, professional or friendly messages.
🌱 Gentle Productivity: Suggest a five-minute starting task and identify what can wait.
🔒 Local-First AI: Run model inference in the browser without requiring a paid AI API key or application backend.
📱 Responsive Design: Use the application on desktop or mobile.

The goal isn't to help people work every minute of the day. It's to reduce mental effort so they can spend less time worrying about tasks and more time living.

Demo

🌐 Live Demo: https://github.com/021abhimanyu/LazyDesk-AI.git

The demo lets you enter your tasks, generate a manageable plan, and draft messages using AI.

Note: The first model download requires an internet connection.

Code

💻 GitHub Repository: https://github.com/021abhimanyu/LazyDesk-AI.git

The project is built using HTML, CSS, and JavaScript. The source code includes the website interface, AI integration, and project documentation.

How I Built It

I wanted to build something that reflects the spirit of the Slow and Steady challenge: a tool that does a little work for you without making productivity feel like another full-time job.

Technology stack

HTML5 for the application structure.
CSS3 for a responsive, calming interface.
JavaScript for application logic and AI interactions.
Qwen2.5 0.5B Instruct as the open-weight language model.
Hugging Face Transformers.js to run model inference directly in the browser.
ONNX Runtime WebAssembly as the inference backend.

How the AI works

When a user submits a brain dump, the application sends the text to the model running in the browser. The model is instructed to produce a short plan with priorities, a five-minute starting point, tasks that can wait, and a kind reminder.

For message drafting, the application uses a different prompt to generate a concise message from the user's notes.

The model and its instructions are integrated into the JavaScript application, so there is no requirement for a paid hosted AI inference API.

The first launch downloads the necessary model and library files. Subsequent offline use depends on browser caching, so fully reliable offline operation is not yet guaranteed.

Why Does Open Innovation Matter?

Open innovation is important because AI-powered tools should be adaptable, transparent, and accessible—not just dependent on expensive APIs.

For LazyDesk AI, using an open-weight model makes several things possible.

1. More control over the AI

The prompts and application logic are visible in the repository. Developers can change the assistant's behavior, adjust its tone, experiment with priorities, and explore compatible models.

2. Better privacy by design

User prompts are processed by the model running in the browser instead of being sent to an application backend or a hosted AI inference API.

The model files must initially be downloaded from their hosting source, but the user's task text does not need to be sent to that source for inference.

3. No per-request AI API bill

There is no paid AI API key or per-request inference charge. The user's device performs the inference, although it still consumes local computing resources and requires an initial download.

4. An opportunity to build in the open

Other developers can inspect the code, suggest improvements, change the prompts, improve offline support, and contribute features.

A small model is not perfect, and local inference can be slower on some devices. However, for a lightweight productivity assistant, this approach offers a useful balance of control, privacy, and accessibility.

To me, open innovation means that developers can build useful AI experiences without giving up control over how those experiences work.

My Agent Session

I haven't recorded a DevRelay agent session yet, so I'm leaving this section out for now.

The Lazy Payoff

LazyDesk AI is designed to reduce the effort of deciding what to do next and writing routine messages.

Instead of staring at a long list of tasks, a user can enter everything in one place and get a small, actionable starting point.

I haven't measured real-world time savings yet, so I won't claim a specific number. My goal is to make the first step easier and reduce the mental overhead of getting started.

Sometimes, being productive means knowing what can wait.

Prize Categories

I'll select the applicable partner categories based on the official challenge rules and the project's eligibility. I won't claim a category without verifying its requirements.

Built with a sloth-sized ambition: make one thing easier, then go rest. 🦥🌿

Hacktoberfest #OpenSource #AI #JavaScript #TransformersJS #Qwen

GreenQuest AI: Turning Screen Time into Green Time with Open-Source AI | #Hacktoberfest

abhimanyu sharma — Fri, 09 Oct 2026 11:08:57 +0000

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

What I Built

🌿 GreenQuest AI — Turn Screen Time into Green Time!

GreenQuest AI is an AI-powered outdoor adventure app that encourages people to take a break from their screens and reconnect with nature.

The idea is simple: instead of using AI to generate more reasons to stay online, what if we used it to inspire people to go outside?

With GreenQuest AI, users can generate personalized outdoor missions based on their preferred activity, available time, and difficulty. Whether it's exploring a nearby park, observing trees, watching birds, discovering cloud patterns, or helping clean up the environment, every mission encourages users to experience the world beyond their screens.

To make the experience more engaging, I added gamification features, including:

🤖 AI-generated missions: Discover fun outdoor challenges.
⏱️ Activity timer: Set aside dedicated time for each adventure.
🏆 XP and levels: Earn experience points by completing missions.
🔥 Streak tracking: Encourage consistent outdoor habits.
🎖️ Achievement badges: Celebrate progress and milestones.
📊 Progress dashboard: Track completed activities and time spent outdoors.
💾 Local progress storage: Keep supported progress data between visits.

GreenQuest AI is designed for students, developers, remote workers, and anyone who spends too much time in front of a screen and wants a fun reason to step outside.

The mission is to make touching grass a habit, not just a meme. 🌱

Demo

🌐 Live Website: https://021abhimanyu.github.io/GreenQuest-AI/

🐙 GitHub Repository: https://github.com/021abhimanyu/GreenQuest-AI

Try generating an outdoor mission, pick a challenge that fits your schedule, and take it outside!

Code

The complete source code is available on GitHub:

021abhimanyu/GreenQuest-AI

The project is built using three main files:

index.html — The application's structure and interface.
style.css — The nature-inspired design and responsive layout.
script.js — AI integration, mission generation, timer, gamification, and progress tracking.

The goal was to keep the project lightweight and easy to understand, so developers can explore the code, customize it, and build upon it.

Contributions and suggestions are welcome!

How I Built It

I built GreenQuest AI using HTML, CSS, JavaScript, Transformers.js, and an open-weight language model.

The main technology choices are:

HTML5 and CSS3: For the responsive, nature-inspired user interface.
JavaScript: For mission generation logic, timers, progress tracking, XP, levels, streaks, and badges.
Transformers.js: To run compatible machine-learning models directly in the browser.
Qwen2.5-0.5B-Instruct: The open-weight language model configured to generate creative outdoor missions.
ONNX Runtime: For browser-based model inference through the Transformers.js ecosystem.
Browser local storage: To preserve supported user progress without requiring a database.

One of my main goals was to make the AI experience accessible without requiring users to install Ollama, configure a separate backend, or obtain a paid inference API key.

The model is loaded through Transformers.js, with browser inference configured to use WebGPU when available and WebAssembly as a fallback.

Here's how the application works:

The user selects their preferred activity, duration, and difficulty.
JavaScript prepares a prompt using those preferences.
The browser-based AI generates an outdoor mission.
The application displays the mission and its details.
The user heads outdoors, starts the timer, and completes the activity.
The app updates supported progress statistics and awards XP and achievements.

I also included fallback mission logic so the application can remain useful when AI generation fails.

The most interesting part of this project is that the AI's purpose is to encourage less screen time and more real-world activity.

Why Does Open Innovation Matter?

Open innovation made this project possible without depending on a proprietary AI inference API.

Using an open-weight model with Transformers.js let me explore browser-based AI inference while keeping the architecture relatively simple. I didn't need to set up a dedicated model server or require users to install a separate AI runtime such as Ollama.

This approach offers several advantages:

🌍 Accessibility

Developers and users can experiment with AI without needing a paid inference API subscription. The model still requires an initial download, and performance depends on the user's device and browser.

🔍 Transparency and experimentation

The application code is available for inspection and modification. Developers can experiment with prompts, compatible models, mission categories, and reward systems.

🛠️ Developer freedom

An open model ecosystem makes it easier to learn about model inference and customize the experience instead of relying entirely on a closed, hosted API.

🔐 More control over inference

With browser-side inference, mission-generation prompts can be processed locally rather than sent to a dedicated inference API. Initial model and library downloads still use external hosting, so this is not a claim that the application makes no network requests.

🌱 AI with a purpose beyond the screen

This is the idea that matters most to me. AI applications often encourage people to spend more time interacting with digital content. GreenQuest AI takes a different approach: it uses AI to suggest activities that encourage people to leave their devices behind and experience nature.

Open innovation made it possible to experiment with this idea using accessible tools and a model I can inspect, customize, and build around.

My Agent Session

I haven't included an agent session link in this submission. If I publish a DevRelay session demonstrating the development process, I'll add it here.

Prize Categories

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

Additional partner prize categories will depend on the challenge's official eligibility requirements. I'll list any additional categories only after confirming that GreenQuest AI meets their criteria.

Built with 🌿, JavaScript, and open-source AI.

Let's use AI to touch grass, not just scroll past it. 🌎💚