DEV Community: Astra5155j

PHANTOM.EXE 👻 — The Spirit in the Machine

Astra5155j — Fri, 30 Oct 2026 17:31:39 +0000

PHANTOM.EXE 👻 — The Spirit in the Machine

Somewhere between the last line of code and the darkness behind the screen, something is waiting to answer.

What if your terminal wasn't just a tool, but a place where digital ghosts could tell stories, decode strange signals, and answer your questions?

For Hacktoberfest Open-Source AI Challenge — Week 4: Ghost in the Machine, I built PHANTOM.EXE — a spooky, interactive, haunted-terminal experience inspired by retro computers, digital folklore, and the idea that something might be living inside the machine.

🕯️ The terminal is open. The spirit is waiting.

👻 What I Built

PHANTOM.EXE is a browser-based haunted terminal that lets users interact with a mysterious digital entity through an atmospheric, retro-inspired interface.

Instead of building another ordinary chatbot UI, I wanted to make the experience feel like opening an old computer terminal and accidentally waking something up.

The experience includes:

Summon a Story — enter the world of eerie digital storytelling.
Ask the Spirit — interact with the mysterious entity.
Decode the Signal — explore the terminal's cryptic side.
Haunted Terminal Aesthetic — a dark interface with atmospheric visual effects inspired by retro computing and horror.
Optional AI Integration — connect the experience to an open-weight language model through Hugging Face Inference Providers.

The goal was to combine the atmosphere of an old-school computer horror story with the accessibility of a web-based AI experiment.

🖥️ Demo

Try PHANTOM.EXE here:

👉 https://huggingface.co/spaces/xcasket/phantom-exe

Open the terminal, explore its interactions, and see what the spirit has to say.

💻 Code

GitHub repository:

👉 https://github.com/Astra5155j/phantom-exe

The project is built with web technologies and designed to be easy to explore, modify, and extend.

If you're interested in contributing, experimenting with the interface, or connecting a different model, the source code is the place to start.

🛠️ How I Built It

I built PHANTOM.EXE using:

HTML for the interface structure.
CSS for the visual design, atmosphere, and terminal styling.
Vanilla JavaScript for interactive behavior.
Hugging Face Spaces to host the live experience.
Hugging Face Inference Providers for optional integration with an open-weight language model.

I wanted the project to remain lightweight instead of requiring a complicated backend or a large infrastructure setup.

The experience supports a scripted demonstration mode, so users can explore its core interactions without configuring an AI provider. An optional live-inference mode can connect to a compatible model, such as "Qwen/Qwen2.5-7B-Instruct", when the required provider access and configuration are available.

This separation makes the interface accessible while leaving room for experimentation with real model inference.

One important implementation detail: the optional inference mode sends requests from the browser to the configured provider. A user-provided access token should never be committed to a public repository. For a production deployment, I would move sensitive credentials to a secure backend rather than expose them in client-side code.

🌐 Why Does Open Innovation Matter?

Open innovation matters because AI should be something people can understand, experiment with, adapt, and build upon — not merely consume through a closed interface.

PHANTOM.EXE is a small project, but it represents several reasons I find open technologies exciting.

Freedom to experiment

Open-weight models give developers the opportunity to explore different models and behaviors instead of being permanently tied to one proprietary model or provider. The model can become a replaceable component rather than the entire product.

Accessibility for independent builders

You don't need a massive team or expensive infrastructure to start building an interesting AI experience. Lightweight web technologies and publicly available model ecosystems give individual developers room to experiment and learn.

More control over the experience

When developers can inspect and modify their code, they can change the interface, the behavior, the prompts, and the way the model is integrated. That flexibility is valuable when you're building something unusual rather than following a standard chatbot template.

A path toward privacy-conscious AI

Open-weight models can enable local inference, which can help keep sensitive data on a user's own machine. PHANTOM.EXE's optional hosted inference path does not automatically provide that guarantee: requests sent to an external provider are subject to that provider's policies. A future local-inference implementation could offer a more private, offline experience.

To me, open innovation isn't simply about getting access to a model. It's about having the freedom to understand the technology, change it, and create something personal with it.

🌼 The Spirit of It

The Ghost in the Machine theme immediately made me think about old computers, forgotten software, abandoned digital spaces, and the strange feeling that something familiar can become unfamiliar after dark.

Technology is full of things we leave behind: old programs, old messages, old projects, and ideas that never quite disappear.

PHANTOM.EXE takes that feeling and turns it into a small interactive experience. The terminal becomes a doorway, the signal becomes a mystery, and the AI becomes a voice emerging from the machine.

I didn't want to build just another tool that answers questions. I wanted to create something with a little personality and atmosphere — something that makes people curious enough to click one more button.

Because sometimes the most interesting part of technology isn't just what it can do.

It's what we imagine might be hiding inside it.

🚀 What's Next?

PHANTOM.EXE is a foundation that can grow beyond its current experience.

Some ideas I'd love to explore:

Local inference for a more private, offline séance.
Additional personalities and narrative paths.
Persistent stories and interactive mysteries.
More terminal commands and hidden interactions.
Support for different open-weight models.

The spirit doesn't have to remain the same forever. That's part of the appeal of building with adaptable technologies.

🕯️ Final Transmission

PHANTOM.EXE started with a simple idea: give a haunted terminal a voice.

It became an experiment in atmosphere, interactive design, and the creative possibilities of open-weight AI.

If you try it, I'd love to know: what would you ask a ghost living inside a computer?

👻 PHANTOM.EXE — Built for the curious, not the cursed.

Built for the Hacktoberfest Open-Source AI Challenge — Week 4: Ghost in the Machine.

devchallenge #hf26challenge #ai #opensource

GrowSight: A Local AI Website Auditor for Small Businesses

Astra5155j — Sat, 17 Oct 2026 23:43:27 +0000

This is a submission for the "Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle" (https://dev.to/challenges/hacktoberfest-week2-2026-10-12).

GrowSight: A Local AI Website Auditor for Small Businesses

A website can look perfectly fine and still lose potential customers.

Maybe the call-to-action is unclear. Maybe the mobile experience needs improvement. Maybe the business is missing a feature that could make it easier for customers to buy, book, or get in touch.

For small businesses, figuring out what to fix can be expensive and time-consuming.

That's why I built GrowSight — an AI-powered website auditing tool that helps businesses identify potential usability issues and discover practical opportunities to improve their websites.

My goal is simple: make useful website analysis more accessible to businesses that cannot afford expensive consultants.

What I Built

GrowSight takes a public website URL and analyzes the page to identify potential improvements.

It examines website structure, content, calls to action, accessibility signals, and desktop and mobile layouts. It then combines automated checks with AI-assisted analysis to turn observations into actionable recommendations.

Some of the things it checks include:

Page titles, metadata, and headings
Buttons, links, and calls to action
Images and missing alternative text
Desktop and mobile website layouts
Potential usability improvements
Features a business might consider adding to improve the customer experience

Rather than simply listing technical problems, GrowSight aims to help answer a more useful question:

What should I improve on my website next, and why?

The tool is designed with freelancers, local businesses, independent creators, and small online stores in mind.

These recommendations are suggestions to investigate, not guaranteed increases in revenue or conversions.

Demo

🎥 Watch the GrowSight demo:

https://drive.google.com/file/d/1Rjo7K7DuMkV2DFnJc6maXphEYbbKa3n3/view?usp=drivesdk

The recording demonstrates my project running locally on my Fedora laptop.

Code

GitHub repository: https://github.com/Astra5155j/growsight-gemma

The repository contains the frontend, backend API, browser-scanning logic, AI integration, and tests.

How I Built It

I wanted to build something useful with open-weight AI rather than just wrap a hosted chatbot in a user interface.

Here's the stack behind GrowSight:

Frontend: HTML, CSS, and vanilla JavaScript
Backend: Python and FastAPI
Browser automation: Playwright
AI model: Gemma 3 4B
Local inference: Ollama
Visual analysis: Website screenshots processed through the vision-capable model when enabled

The workflow starts with a URL. Playwright opens the website and collects information about its structure and appearance, including desktop and mobile screenshots. Rule-based checks identify straightforward issues, while the AI integration uses the collected evidence to help generate recommendations.

The application is designed to run on a personal computer instead of requiring a dedicated cloud AI service.

This is an early MVP, and I still have work to do on scan reliability, recommendation quality, and testing the usefulness of the suggestions with real business owners.

Why Does Open Innovation Matter?

For a tool aimed at small businesses, the economics matter.

If every audit requires a paid AI API call, recurring costs can make experimentation and long-term use more difficult. Depending entirely on one provider can also limit how a project evolves.

Running Gemma locally gives GrowSight a different starting point.

Lower marginal AI inference costs

Once the model is downloaded, local inference does not require a separate paid AI API for every request. Hardware, electricity, and maintenance still have costs, but there is no per-request fee charged by an external model API.

More control over data

The AI inference can happen on the user's own machine instead of sending collected audit evidence to a third-party AI API. The website still needs to be accessed to scan a live public URL, but the AI component can remain local.

Freedom to experiment

I can change prompts, modify the auditing logic, and explore different compatible models without rebuilding the entire application around a proprietary AI provider.

A path toward offline AI

The model can perform inference locally without an external AI API. Scanning a live website still requires internet access, so the entire workflow is not fully offline.

For me, open innovation is not just about using a free model. It is about having the freedom to experiment, understand the system, and build a useful product without making a closed AI service the centre of everything.

What I Learned

Building GrowSight made me think about the difference between a generic AI answer and an AI system grounded in actual evidence.

A model can produce convincing recommendations, but useful website analysis needs context: what the page contains, how it is structured, what users can click, and how the layout behaves on different screens.

Combining browser automation, deterministic checks, and local AI gives me a foundation to explore that problem.

It also made one thing clear: generating recommendations is only the beginning. The real test is whether a business owner finds them useful enough to act on.

What's Next?

My next priorities are to improve scan reliability, make recommendations more specific to different business types, improve visual comparisons, and get feedback from real website owners.

Eventually, I want GrowSight to become a practical tool that helps a small business owner identify what deserves attention first, rather than leaving them with a long list of generic suggestions.

I'm building toward a simple idea:

Your website should work for your business, not just exist online.

My First User

I am looking to get GrowSight in front of small business owners and learn which recommendations are genuinely useful to them.

I haven't included any claims about paying customers or measured revenue improvements because those need to come from real usage.

If you run a small business or know someone who does, I'd love to hear what you would want a tool like this to identify on your website.

RoomMate AI — AI That Actually Uses Your Room

Astra5155j — Mon, 05 Oct 2026 00:25:39 +0000

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

What I Built

I built **
RoomMate AI** for someone who is very dear and close to me.

She had recently shifted to a hostel for a new college, and one night her new room was feeling especially gloomy. She was having a difficult time with the way the room looked and felt, and she wanted to make it feel more like her own space.

She had been trying to use AI to generate ideas for her favourite kind of room, but there was a problem: most AI room generators would create a completely new room instead of actually working with the room she already had.

That's what got me thinking:

What if AI could actually look at her real room and help her figure out what would work there?

So I ended up building RoomMate.

You upload a photo of your actual room, describe the style and purpose you want, and set a budget. RoomMate uses AI to look at the room and suggest decorations and changes that could work with the space and furniture that are already there.

But I didn't want it to stop at recommendations.

You can actually take those recommendations and place them into the room using the interactive editor. You can move, resize and rotate the decorations and see how the room could look before deciding what you actually want to change.

Instead of:

"Here is an AI-generated room that looks nothing like mine."

RoomMate tries to answer:

"Here is what your room could look like using the space you already have."

Demo

Watch the RoomMate AI demo

The demo shows the complete flow:


text
Upload room
    ↓
Choose style + purpose + budget
    ↓
Gemma 3 analyzes the room
    ↓
Get personalized recommendations
    ↓
Add recommendations to the room
    ↓
Move / resize / rotate decorations
    ↓
See the redesigned room