DEV Community: Cyber

GhostTrace — The Shell Remembers Hacktoberfest 2026 Week 4 cover is ready. 👻.

Cyber — Sun, 01 Nov 2026 05:29:04 +0000

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

What I Built

👻 GhostTrace — The Shell Remembers

GhostTrace is a privacy-first, AI-powered haunted terminal companion that brings a supernatural personality to the command line.

Inspired by the concept of a ghost living inside a computer, I wanted to build something that doesn't just execute commands but also responds to them with its own personality.

Key Features:

AI-Powered Whispers: Generates spooky, context-aware commentary based on terminal commands and their outcomes.
Local AI Inference: Uses Google's open-weight Gemma 3 4B model through Ollama.
Command Safety: Validates commands and blocks potentially dangerous operations before execution.
Multiple Ghost Personas: Includes GhostTrace, Spectral Wraith, Cyber Poltergeist, and Crypt Keeper.
Haunt.sh Integration: Extends the experience into interactive Bash shells with asynchronous ghost whispers.
Offline Fallback: Provides deterministic responses when the local AI model is unavailable.
Privacy-First Architecture: Designed to keep AI inference local without depending on cloud AI APIs or cloud telemetry.

The idea is simple: What if your terminal had a ghost living inside it?

Code

GitHub Repository:https://github.com/Suguda-Thakur-Marndi/GhostTrace

The repository includes the Python terminal companion, Bash shell integration, local AI daemon, ghost personas, security controls, automated tests, and documentation.

How I Built It

I built GhostTrace using open-source technologies and locally running AI models.

AI Model:

Google Gemma 3 4B through Ollama.
Optional llama.cpp provider support for local inference.
Deterministic fallback responses for offline operation.

Tech Stack:

Python 3.10+
Bash
Zsh integration
Ollama
Gemma 3 4B
llama.cpp
Pytest

Architecture:

The project has two main components:

GhostTrace Python REPL: An interactive terminal companion responsible for command execution, safety validation, local AI integration, and context-aware commentary.
Haunt.sh: A shell integration that hooks into interactive Bash sessions and communicates with a local Python daemon to generate asynchronous ghost whispers without blocking the shell prompt.

Security and privacy are central to the architecture. Commands are validated, sensitive information is redacted, and AI-generated responses are sanitized rather than executed.

The project has also been tested with an automated Python test suite, with 287 tests passing in the latest fully successful run.

Why Does Open Innovation Matter?

Open innovation made GhostTrace possible.

Using open-weight models gives developers the freedom to run AI locally, customize model behavior, experiment with different models, and build applications without depending entirely on proprietary cloud APIs.

For GhostTrace, this is particularly important because terminal interactions can contain sensitive information. Local inference helps keep that processing under the user's control.

Open-source development also allows contributors to inspect the implementation, audit security mechanisms, create new ghost personas, improve the command safety system, and extend the project.

GhostTrace demonstrates how open-source AI can make developer tools more creative, accessible, customizable, and privacy-conscious.

My Agent Session

Optional: [Add your DevRelay agent session link here]

The Spirit of It

Imagine opening your terminal late at night.

You type:

whoami

The command executes successfully.

Then, somewhere inside the machine, a voice responds:

"Another presence has entered this system. I remember you."

That is the experience I wanted to create.

I wanted to combine practical developer tooling with a little mystery, nostalgia, and supernatural personality.

A terminal that feels alive. A machine that responds. A ghost that never sleeps.

The shell remembers. The ghost never sleeps. 👻

Prize Categories

Featured Category

Gemma — Best Use of Gemma ($200): Uses Google's open-weight Gemma 3 4B model through Ollama for locally generated, context-aware terminal commentary and privacy-focused AI interactions.

🥦 FridgeFlow AI: Your Smart Kitchen Assistant | Hacktoberfest Week 3 #devchallenge #hf26challenge #OpenSource #AI

Cyber — Fri, 23 Oct 2026 17:34:47 +0000

🥦 FridgeFlow AI — Your Smart Kitchen Assistant | Hacktoberfest Week 3

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

What I Built

FridgeFlow AI is a smart food management application designed to simplify everyday cooking and help reduce food waste.

Instead of manually tracking everything in your fridge, FridgeFlow AI aims to help users:

📸 Scan food ingredients and manage fridge inventory.
⏰ Track expiry dates and identify food that needs attention.
🍳 Discover recipes based on available ingredients.
🛒 Create and manage grocery lists.
🌱 Reduce food waste and make smarter use of groceries.

The goal is simple: spend less time managing your kitchen and more time enjoying your food.

Demo

🎬 Watch the YouTube Demo: https://youtu.be/DAs7VaVvlYo

🌐 Live Application: https://fridgeflow-ai-mobile.vercel.app

⚙️ Backend API: https://fridgeflow-ai-api.onrender.com

Code

💻 GitHub Repository: https://github.com/Suguda-Thakur-Marndi/FridgeFlow-AI

How I Built It

FridgeFlow AI combines cross-platform application development with a Python backend and AI-powered food-management capabilities.

Tech stack:

React Native + Expo — cross-platform mobile app development.
TypeScript — maintainable, type-safe application code.
FastAPI + Python — backend APIs and application logic.
Expo SQLite — local storage for inventory and grocery data.
Docker — containerizing the backend for deployment.
Render — hosting the FastAPI backend.
Vercel — planned hosting for the Expo web frontend.
Gemma 3 4B — planned integration for local food-ingredient recognition and recipe assistance, pending verified implementation.

A major engineering goal is to explore privacy-friendly, offline-first AI while making food management accessible across devices.

Why Does Open Innovation Matter?

Open innovation makes it possible to build, inspect, customize, and improve software using technologies developed by the community.

For FridgeFlow AI, open-source frameworks provide flexibility over the application architecture and create opportunities to experiment with local AI inference instead of depending entirely on closed AI services.

Open-source development also makes it easier for other developers to contribute improvements, experiment with new features, and extend the project.

My Agent Session

🤖 Agent Session: Add your DevRelay session link here if available.

The Lazy Payoff

Imagine opening your fridge, checking what you already have, discovering a recipe, and preparing your grocery list without juggling multiple apps.

FridgeFlow AI aims to make that process simpler: less manual tracking, less food waste, and easier everyday cooking.

Prize Categories

Gemma — Best Use of Gemma : Planned integration of Gemma 3 4B for local food-ingredient recognition and recipe assistance, enabling privacy-focused AI inference on supported devices.
Render — Best Use of Render : Deploying the FridgeFlow AI FastAPI backend on Render to support food-inventory management, recipe generation, and AI-powered kitchen assistance.

These entries depend on demonstrating the required integrations and working functionality. Gemma inference must be verified before claiming it as an implemented feature.

Building FridgeFlow AI for Hacktoberfest 2026 Week 3: Slow and Steady 🥦

devchallenge #hf26challenge #Hacktoberfest #OpenSource #AI #Gemma #Render #ReactNative #FastAPI #FoodTech #Sustainability #FridgeFlowAI

LocalLedger AI: A Privacy-First Receipt & Expense Manager #devchallenge #hf26challenge #OpenSource #AI

Cyber — Thu, 15 Oct 2026 20:17:53 +0000

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

What I Built

I built LocalLedger AI, a privacy-focused Windows desktop application designed to help freelancers, individuals, and small business owners manage receipts, expenses, and financial records.

Managing financial records manually can be time-consuming. LocalLedger AI aims to simplify this process through receipt processing, document extraction, transaction management, and financial reporting.

Key Features

🧾 Receipt and invoice processing using OCR and PDF extraction
💰 Expense and transaction management
📊 Financial reports and expense insights
🔒 Local-first record storage using SQLite
🤖 Optional local AI integration using Ollama and Qwen 2.5 1.5B
🖥️ Windows desktop application built with Electron

The project could be useful to freelancers, independent professionals, and small businesses looking for a straightforward way to organize financial records.

Demo

🎥 YouTube Demo: https://youtu.be/b8UtQg6KkzU

📥 Download the Windows Installer:

https://github.com/Suguda-Thakur-Marndi/LocalLedger-AI/releases/download/v1.0.0/LocalLedger.AI.exe

Code

💻 GitHub Repository:

https://github.com/Suguda-Thakur-Marndi/LocalLedger-AI

The project is built with open-source technologies and is available for others to explore, learn from, and contribute to.

How I Built It

I built LocalLedger AI using a combination of web technologies, desktop application tooling, document-processing libraries, and local AI.

Tech stack:

Frontend: React, TypeScript, Vite
Desktop application: Electron
Backend: Python, FastAPI
Database: SQLite
OCR and document processing: Tesseract OCR, PyMuPDF
Local AI: Ollama with Qwen 2.5 1.5B

The application combines a desktop interface with a Python backend for financial record management and document processing. Ollama provides an option to run supported AI workflows locally rather than depending on a hosted AI API.

My goal was to explore how local AI and open-source tools can make everyday financial workflows more accessible while supporting user privacy.

Why Does Open Innovation Matter?

Open innovation matters because useful software should be possible to build without depending entirely on closed platforms or paid AI APIs.

For LocalLedger AI, open-source technologies made it possible to combine document processing, local data storage, desktop development, and local inference into one application.

Using local AI tools such as Ollama and an open-weight model such as Qwen 2.5 1.5B also gave me an opportunity to explore AI workflows without requiring a hosted model API for those local inference tasks.

Open innovation allows developers to inspect implementations, adapt tools to their needs, learn from existing projects, and collaborate with the wider community.

For a financial application, local-first design is particularly interesting because financial documents can contain sensitive information. Local processing can reduce the need to send those documents to external AI services, although the privacy guarantees depend on how each workflow is configured.

Prize Categories

GitHub — Best Use of GitHub Copilot

I'm exploring ways to improve LocalLedger AI's development workflow through automated quality checks, GitHub Actions, and pull-request reviews.

The intended workflow includes frontend and backend testing, automated checks on pull requests, and reviewing proposed code changes before merging.

This category entry should be retained only if I use and document qualifying GitHub Copilot functionality—such as its coding agent, CLI, app, or pull-request review features—in accordance with the official challenge rules. GitHub Actions alone or the use of another AI coding assistant does not establish eligibility.

Built for Hacktoberfest Week 2: Side Hustle. 🚀

Feedback, suggestions, and contributions are welcome!

WildLens AI: Discover Nature with Open-Source AI 🌿 | Hacktoberfest 2026

Cyber — Sun, 11 Oct 2026 13:40:43 +0000

🌿 WildLens AI: Open-Source AI That Gets You Outside

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

What I Built

WildLens AI is an AI-powered nature exploration application designed to turn everyday outdoor walks into opportunities for discovery.

Have you ever spotted an interesting flower, plant, or unfamiliar species while walking outside and wondered what it was?

WildLens AI helps users explore that curiosity by allowing them to capture or upload an image and discover possible species matches, including scientific names and relevant information.

The project aims to make nature exploration more interactive and accessible for students, nature enthusiasts, hikers, gardeners, and anyone curious about the biodiversity around them.

🌱 The idea is simple: use AI as a reason to step away from the screen, explore the outdoors, and learn something about the natural world.

Beyond identification, WildLens AI includes features such as Field Missions and a Field Journal to encourage users to observe, discover, and document their experiences.

Demo

🌐 Live Website: WildLens AI

🎥 Video Demo: Add your published video link here.

The frontend is deployed on GitHub Pages. Production AI identification depends on a separately hosted and correctly configured backend.

Code

💻 GitHub Repository: WildLens AI

The project is being developed with an emphasis on open-source AI, modular architecture, and a practical nature-exploration experience.

Contributions, feedback, bug reports, and ideas for improving the project are welcome!

How I Built It

WildLens AI combines a web application, an AI-powered identification pipeline, and nature-focused exploration features.

Open-Source AI

The core identification model is BioCLIP-2, an open-weight vision-language model developed for biological imagery.

Using the open_clip library, the identification pipeline encodes an image and compares its visual representation against text representations of candidate species. Similarity scores help rank possible matches.

These scores are not calibrated probabilities, and the highest-ranked candidate is not necessarily correct. Reliable identification requires appropriate candidate coverage, evaluation, and honest communication of uncertainty.

Application Architecture

The project combines:

Frontend: A web interface for image-based exploration and nature-related features.
Backend: Python and FastAPI for server-side functionality and AI inference.
AI model: BioCLIP-2 for image–text similarity and candidate ranking.
Knowledge base: Curated taxonomy information to support species-related information.
Nature exploration: Field Missions and a Field Journal to encourage observation beyond the screen.

The frontend and backend have separate responsibilities, allowing the user interface and inference pipeline to be maintained independently.

GitHub Pages serves the static frontend; live AI identification requires a separately reachable backend.

Why Does Open Innovation Matter?

Biodiversity is enormous, and access to biological knowledge should not be limited to researchers with expensive tools or proprietary AI services.

Open innovation makes projects like WildLens AI possible through open-weight models, reusable libraries, public model documentation, and community-driven development.

Using BioCLIP-2 provides a foundation for experimenting with biological image–text matching without building a vision model from scratch.

Open-source development also makes the project easier for others to inspect, learn from, test, improve, and extend. Contributors can help improve the taxonomy knowledge base, evaluation datasets, user experience, performance, and support for additional species.

For me, open innovation is not just about making code available. It is about making experimentation, learning, and collaboration more accessible.

Prize Categories

Include a partner category here only if your project meets its official eligibility requirements.

Prize Category

*Render — Best Use of Render *

WildLens AI uses a Python FastAPI backend to support AI-powered nature identification and educational features. I plan to use Render to deploy and host the backend API, connecting it to the GitHub Pages frontend. The application combines BioCLIP-2 for biological image identification with Gemini for online nature explanations and question answering.

The goal is to build a reliable, accessible nature exploration platform with clear online/offline functionality and a smooth user experience.

The goal is not to spend more time staring at a screen. It is to use technology to notice the world around us, identify something unfamiliar, and become curious enough to go outside again.

Identify something. Learn something. Look up. Touch grass. 🌿