DEV Community: Devansh Shukla

SlothDesk AI — Your Local-First File Organization Assistant

Devansh Shukla — Sun, 25 Oct 2026 21:28:47 +0000

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

What I Built

Meet SlothDesk AI 🦥 — a local-first file organization assistant that helps you spend less time sorting files and more time doing what actually matters.

Downloads folders have a habit of becoming digital junk drawers. Documents, images, archives, and miscellaneous files pile up until finding anything becomes a task of its own.

SlothDesk AI helps bring order to that chaos.

It scans a folder, analyzes file metadata, and builds an organization plan that you can review before applying any changes. It can also use optional, locally running Ollama models to suggest file categories and more meaningful filenames.

The principle behind it is simple: AI should do the tedious thinking, but you should stay in control.

What makes it different?

🌿 Local-first: Work with files on your own device.

🧠 Optional local AI: Get organization suggestions through Ollama without depending on a paid AI API.

🛡️ Human approval: Review the proposed changes before applying them.

📁 Safer organization: Designed to avoid accidental overwrites and file deletion.

⚡ Simple workflow: Scan, review, and organize through a Streamlit interface.

It's for anyone whose Downloads folder needs a little less chaos — students, developers, researchers, and anyone who would rather not sort files manually.

Demo

🎥 Demo video: Watch my SlothDesk AI walkthrough

SlothDesk AI.mp4 - Google Drive

drive.google.com

The walkthrough demonstrates the app's interface and file-organization workflow.

Code

💻 GitHub repository:
https://github.com/devanshshukla-3004/SlothDesk-AI-Your-local-AI-productivity-assistant

The project is built as an open-source Python application. Explore the repository for the implementation, setup instructions, and tests.

How I Built It

SlothDesk AI combines a lightweight Python application with an optional local AI suggestion engine.

Tech stack:

Python — core application and file-organization logic

Streamlit — interactive user interface

Pandas — tabular presentation of scanned file information

Ollama — optional local inference for AI-powered organization suggestions

Pytest — automated testing

The workflow separates planning from execution. The application scans the selected folder and builds an organization plan first. The user can review that plan before choosing to apply it.

The AI component is optional. When enabled, it helps suggest categories and filenames using file metadata rather than sending file contents to the suggestion engine. This keeps the project focused on practical assistance without giving an AI model unrestricted control over the filesystem.

I wanted the experience to feel useful even without a cloud AI subscription. Local inference makes that possible for users with a compatible Ollama setup and model.

Why Does Open Innovation Matter?

For a project like SlothDesk AI, open innovation isn't just about making the source code visible. It changes who can use, inspect, adapt, and improve the tool.

Using open-source tools and local inference makes experimentation possible without requiring a paid proprietary AI API. Developers can inspect the implementation, run it in their own environment, test different compatible models, and adapt the workflow to their needs.

It also makes privacy-conscious experimentation more accessible. Instead of building the entire experience around sending data to a remote service, SlothDesk AI can use local models for its optional suggestions.

There is still work to do: local inference depends on the user's hardware and installed model, and AI-generated suggestions should always be reviewed. But open tools make it possible to build and improve this kind of practical assistant with fewer external dependencies.

That's the kind of open innovation I wanted to explore: use AI to reduce repetitive work without sacrificing user control.

The Lazy Payoff

The goal is to reduce the repetitive effort involved in sorting a cluttered folder.

Rather than manually deciding where each file belongs and thinking up better names one at a time, users can scan a folder, review suggested organization, and approve a plan.

I haven't measured a verified time-saving figure yet, so I won't claim a specific number of minutes saved. The next step is to test it on real, cluttered folders and measure how much manual effort the workflow actually removes.

Slow and steady — build something useful, test it honestly, and keep improving it. 🌱

Prize Categories

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

SlothDesk AI uses optional local inference through Ollama to support privacy-conscious file organization without requiring a paid proprietary AI API.

PhishBrief AI: Local-First Phishing Analysis with Open-Weight AI

Devansh Shukla — Tue, 13 Oct 2026 12:43:10 +0000

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

What I Built

PhishBrief AI is a local-first AI-powered phishing analysis tool designed to help small businesses and individuals understand suspicious emails and messages before interacting with them.

Phishing emails often use urgency, impersonation, suspicious links, and requests for sensitive information to trick people. Many small teams lack dedicated security staff or enterprise-grade email security tools.

PhishBrief AI aims to make initial phishing analysis more accessible by helping users examine suspicious content, understand potential warning signs, and receive actionable security recommendations.

The project explores a privacy-conscious approach: running an open-weight AI model locally instead of requiring every message to be sent to a hosted AI API.

Who could benefit?

Small businesses without a dedicated security operations team

Freelancers and independent professionals

Students learning email security and phishing detection

IT support teams that need an accessible first-pass analysis tool

The long-term side-hustle opportunity is a lightweight security assistant for small teams that need understandable phishing triage without the cost or complexity of a full enterprise security platform.

Demo

🎥 Watch the demo: https://www.loom.com/share/ca8e3321c0274ec49f58d0d423f2813b

The demo shows the application running locally and demonstrates its phishing-analysis workflow.

Code

💻 GitHub repository: https://github.com/devanshshukla-3004/PhishBrief-AI

Explore the source code, setup instructions, and project implementation in the repository.

How I Built It

PhishBrief AI combines a Python application with local AI inference through Ollama.

Core technologies

Python — application logic and analysis workflow

Streamlit — interactive user interface

Ollama — local model runtime

Granite 3.2 Vision — the configured open-weight model used for local inference

The development setup also includes Gemma 3 1B as another locally available model.

The workflow is designed around analyzing user-provided suspicious content, identifying potential phishing indicators, explaining the associated risks, and presenting recommendations in a more understandable format.

The emphasis is on making AI-assisted security analysis accessible while exploring local inference as an alternative to sending potentially sensitive content to a third-party AI API.

This is an assistive analysis tool, not a guarantee that every phishing attempt will be detected. Users should still verify suspicious messages and links independently.

Why Does Open Innovation Matter?

Open innovation makes projects like PhishBrief AI possible for developers who do not have access to expensive infrastructure or enterprise security systems.

Open-weight models and tools such as Ollama make it possible to experiment with local inference, learn how AI systems behave, and build practical applications around them.

For this project, local inference offers three important advantages:

Privacy-conscious experimentation: Potentially sensitive email content can be processed locally, depending on the application's implementation and configuration.

Lower barriers to development: Developers can experiment without depending entirely on paid, hosted AI APIs.

Customisation and learning: Open-weight models make it easier to explore model choices, prompting strategies, and application workflows.

Open innovation does not automatically guarantee security or accuracy, but it gives independent developers more control over how AI-powered tools are built and evaluated.

My First User

My first target users for PhishBrief AI are individuals and small-business teams who need a simple way to understand suspicious emails without requiring advanced cybersecurity knowledge.

I'm looking for early testers who can try the tool with safe sample phishing emails, evaluate whether the risk explanations are understandable, and share feedback on its usefulness and ease of use.

The goal is to learn whether local AI-powered phishing analysis can make basic security triage more accessible to people without a dedicated security team.

I'm currently looking to gather feedback from early users and improve the project based on their experience.

Prize Categories

PhishBrief AI is my submission to the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle.

The project focuses on building a practical AI-powered tool around open-weight models and local inference.

What's Next?

Some areas I want to explore next include:

Improving phishing-indicator extraction and explanation quality

Testing the tool against a broader collection of safe, labelled examples

Adding more structured risk reporting

Improving evaluation, error handling, and documentation

Making the setup easier for non-technical users

Building PhishBrief AI has been an opportunity to explore how open-weight AI can support practical cybersecurity workflows.

I'd love to hear feedback from developers, security researchers, and anyone experimenting with local AI.

devchallengesvg #hf26challengesvg #cybersecurity #ai

PhishBrief AI: Private, Local-First Phishing Analysis with Open-Weight AI

Devansh Shukla — Tue, 13 Oct 2026 12:39:33 +0000

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

What I Built

PhishBrief AI is a local-first AI-powered phishing analysis tool designed to help small businesses and individuals understand suspicious emails and messages before interacting with them.

PhishBrief AI aims to make initial phishing analysis more accessible by helping users examine suspicious content, understand potential warning signs, and receive actionable security recommendations.

The project explores a privacy-conscious approach: running an open-weight AI model locally instead of requiring every message to be sent to a hosted AI API.

Who could benefit?

Small businesses without a dedicated security operations team

Freelancers and independent professionals

Students learning email security and phishing detection

IT support teams that need an accessible first-pass analysis tool

Demo

Watch the demo: https://www.loom.com/share/ca8e3321c0274ec49f58d0d423f2813b

The demo shows the application running locally and demonstrates its phishing-analysis workflow.

Code

GitHub repository:
https://github.com/devanshshukla-3004/PhishBrief-AI

Explore the source code, setup instructions, and project implementation in the repository.

How I Built It

PhishBrief AI combines a Python application with local AI inference through Ollama.

Core technologies

Python — application logic and analysis workflow

Streamlit — interactive user interface

Ollama — local model runtime

Granite 3.2 Vision — the configured open-weight model used for local inference

The development setup also includes Gemma 3 1B as another locally available model.

The emphasis is on making AI-assisted security analysis accessible while exploring local inference as an alternative to sending potentially sensitive content to a third-party AI API.

This is an assistive analysis tool, not a guarantee that every phishing attempt will be detected. Users should still verify suspicious messages and links independently.
**

Why Does Open Innovation Matter?**

Open innovation makes projects like PhishBrief AI possible for developers who do not have access to expensive infrastructure or enterprise security systems.

Open-weight models and tools such as Ollama make it possible to experiment with local inference, learn how AI systems behave, and build practical applications around them.

For this project, local inference offers three important advantages:

Privacy-conscious experimentation: Potentially sensitive email content can be processed locally, depending on the application's implementation and configuration.

Lower barriers to development: Developers can experiment without depending entirely on paid, hosted AI APIs.

Customisation and learning: Open-weight models make it easier to explore model choices, prompting strategies, and application workflows.

Open innovation does not automatically guarantee security or accuracy, but it gives independent developers more control over how AI-powered tools are built and evaluated.

My First User

My first target users for PhishBrief AI are individuals and small-business teams who need a simple way to understand suspicious emails without requiring advanced cybersecurity knowledge.

I'm looking for early testers who can try the tool with safe sample phishing emails, evaluate whether the risk explanations are understandable, and share feedback on its usefulness and ease of use.

The goal is to learn whether local AI-powered phishing analysis can make basic security triage more accessible to people without a dedicated security team.

I'm currently looking to gather feedback from early users and improve the project based on their experience.

Prize Categories

PhishBrief AI is my submission to the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle.

The project focuses on building a practical AI-powered tool around open-weight models and local inference.

🌿 WildHunt AI — Turn AI Into a Reason to Go Outside

Devansh Shukla — Wed, 07 Oct 2026 08:08:40 +0000

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

What I Built

WildHunt AI is a privacy-first, open-source AI scavenger hunt that turns your phone into a guide for exploring the physical world.

Instead of asking people to spend more time interacting with an AI chatbot, WildHunt gives them a reason to put the screen down, go outside, observe their surroundings, and come back only when they have something worth showing the AI.

The core loop is:

Choose a Hunt
↓
Get a mission
↓
Go outside
↓
Find the target
↓
Take a photo
↓
Local AI verifies it
↓
Unlock the next discovery

WildHunt supports different exploration styles such as:

🌿 Nature
🏙️ Urban Explorer
🎨 Colors
🪨 Texture
👀 Observation
📍 Landmark Hunt

The important part is that the AI isn't simply performing generic image classification.

It evaluates the submitted photo against the specific mission the user was given.

For example, if the mission is to find a naturally yellow object outdoors, the AI checks whether the photograph actually satisfies that mission.

Most AI products compete for your attention. WildHunt uses AI to give your attention back to the physical world.

Who is it for?

WildHunt is designed for anyone who wants a lightweight reason to explore:

-Students taking a break from studying
-Developers spending too much time at their desks
-Families looking for simple outdoor activities
-People exploring a new neighborhood
-Anyone who wants to turn a walk into a small adventure

Demo

🎥 Video Demo: https://drive.google.com/file/d/17hSBEzV7f4hfPKQCj0GlNb3MraOQQwBd/view?usp=sharing

The demo shows the complete WildHunt-AI experience:

1.Starting a hunt
2.Receiving a mission
3.Capturing a discovery
4.Sending the image to the local vision model
5.Receiving mission-specific verification
6.Continuing the hunt

The application runs as a responsive Progressive Web App, so it can be used directly from a phone browser without requiring a native mobile application.

Code

🔗 GitHub:

devanshshukla-3004 / WildHunt-AI

🌿 WildHunt AI

The screen gives you the mission. The real world gives you the answers.

WildHunt AI is a privacy-first outdoor scavenger-hunt PWA that uses local open-weight vision AI to turn real-world exploration into an interactive game.

Choose a hunt. Get a mission. Go outside. Photograph your discovery. Your own machine's vision model checks whether the image satisfies the mission — without sending the photo to a WildHunt cloud backend.

✦ Why WildHunt?

Most AI products compete for your attention.

WildHunt uses AI to give your attention back to the physical world.

The product is intentionally designed around a simple loop:

Choose Hunt
    ↓
Receive Mission
    ↓
Go Outside
    ↓
Find Something Real
    ↓
Take a Photo
    ↓
Local Vision AI Verifies It
    ↓
Discovery Unlocked
    ↓
Next Mission

The screen is only the starting point. The real world is the game board.

🎥 Product Walkthrough

WildHunt is designed…

View on GitHub

WildHunt is fully open source.

The repository includes the application source, AI integration, safety logic, documentation, tests, contribution guidelines, and project architecture.

How I Built It

WildHunt is built around local open-source AI rather than a paid proprietary AI API.

Tech Stack
React + TypeScript
Vite
Progressive Web App
Ollama
Granite 3.2 Vision
Zod
IndexedDB / browser storage
Browser camera APIs
Vitest

The application communicates with a locally running Ollama instance:

WildHunt PWA
│
│ Mission + Image
▼
Local Ollama
│
▼
Granite 3.2 Vision
│
│ Structured verification
▼
WildHunt

The verification response is validated using a schema similar to:

{
"matched": true,
"confidence": 0.91,
"evidence": [
"yellow flower",
"outdoor vegetation"
],
"explanation": "The image clearly shows a yellow flower outdoors."
}
The AI receives the actual mission context, so verification is based on whether the image satisfies that particular challenge.

Privacy by design

There is no requirement for:

Paid AI APIs
Cloud image storage
A database server
User accounts
Analytics
Tracking

The goal is to keep the experience as local and private as possible.

The architecture also keeps the AI provider behind an abstraction layer, making it possible to add other local or browser-based vision providers in the future.

Why Does Open Innovation Matter?

Open innovation is especially important for a project like WildHunt-AI because the entire idea is about changing the relationship between people and AI.

A closed AI API could have made the image verification feature easier to implement, but it would also introduce dependency on a proprietary service, ongoing API costs, and potentially require sending personal photographs to an external provider.

Using open-source/open-weight technology made a different architecture possible:

The AI can run locally.

That means a user's outdoor discoveries don't have to become somebody else's cloud dataset just to determine whether they completed a scavenger-hunt mission.

It also makes the project more accessible to contributors.

Developers can inspect the AI integration, experiment with different local vision models, improve verification logic, add new hunt types, and build alternative inference providers without being locked into a single commercial API.

For me, open innovation isn't just about making the code public.

It's about making the architecture replaceable, inspectable, and hackable.

Prize Categories

Best Use of Entire

WildHunt was developed with an agent-assisted workflow, and the development process is documented through the project's agent session.

🔐 RepoShield AI — Local-First Secret Exposure Detection with Gitleaks + Gemma

Devansh Shukla — Mon, 05 Oct 2026 06:16:20 +0000

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

What I Built

I built RepoShield AI, a local-first security tool that helps developers identify and understand potential secret exposure in their repositories.

I built it for a developer friend who wanted a simpler way to investigate accidentally exposed credentials without sending sensitive repository information to a cloud AI service.

The core idea is simple:
Detect with deterministic security tooling. Explain locally. Never send raw secrets to the AI layer.

RepoShield combines:

🔐 Gitleaks for deterministic secret detection
🛡️ A sanitization layer that withholds raw secrets and absolute paths
🤖 Ollama + Gemma 3 1B for local AI-assisted analysis
📊 A React + TypeScript security dashboard
📄 Sanitized JSON and Markdown reports

The AI model is not responsible for detecting the secret.
Instead:
Repository
↓
Gitleaks
↓
Finding
↓
Secure Sanitization
↓
Safe Metadata
↓
Ollama + Gemma 3 1B
↓
Risk Explanation + Remediation.

This separation was one of the most important security decisions in the project.

Demo

The project has been validated as a local application with a real end-to-end workflow.

The validated demo performs:

drive.google.com

Gitleaks detection
Authorized local repository scanning
Secret/path sanitization
Finding investigation
Local Gemma analysis
Risk explanation
Remediation guidance
Sanitized report generation.

Code

The complete project is open source:

devanshshukla-3004 / RepoShield-AI

Local-first secret exposure review with Gitleaks detection, secure finding sanitization, and optional Ollama/Gemma AI triage.

RepoShield AI

Local-first secret exposure review for repositories — deterministic detection, privacy-preserving triage, and optional local AI assistance.

RepoShield AI helps developers review repositories for potential secret exposure before code is shared. It combines deterministic Gitleaks detection with a deliberately narrow privacy boundary and optional local Ollama / gemma3:1b analysis.

The core design principle is simple:

Detect with deterministic security tooling. Explain locally. Never send raw secrets to the AI layer.

Why RepoShield?

Secret scanners are good at finding patterns. Developers still need to understand what a finding means, how serious it may be, and what to do next.

RepoShield separates those responsibilities:

Gitleaks detects potential secrets.
RepoShield sanitizes the result before it reaches the UI or report layer.
Gemma can explain the finding locally using only server-defined metadata.
The developer triages and remediates the exposure.

This keeps AI in an advisory role rather than allowing a language model…

View on GitHub

The repository contains the React frontend, Express API, security scanning logic, sanitization layer, AI integration, tests, CI configuration, architecture documentation, and demo media.

How I Built It

The project is built around Gemma 3 1B, running locally through Ollama.

The architecture uses:

React
TypeScript
Vite
Tailwind CSS
Express
Zod
Gitleaks
Ollama
Gemma 3 1B
GitHub Actions

The security pipeline intentionally separates detection from AI analysis.

Gitleaks detects potential secrets first. RepoShield then sanitizes the resulting finding before anything reaches the model.

The AI layer receives only controlled finding metadata needed to generate an explanation and remediation guidance.

It does not receive:

❌ Raw secret values
❌ Source code
❌ Absolute filesystem paths
❌ Raw scanner evidence

I also added automated tests around path safety and finding sanitization to make the security boundary testable rather than relying only on documentation.

Why Does Open Innovation Matter?

Open innovation made it possible to build the AI portion of this project around a local, open-weight model rather than requiring a closed cloud AI API.

That matters particularly for cybersecurity.

Security findings can contain sensitive information. Sending those findings to a third-party API creates another trust boundary.

With Ollama and Gemma, I could experiment with a local inference workflow where the AI analysis happens on the developer's machine.

More importantly, open models allowed me to design the system around the question:
What should the AI never receive?

Instead of simply sending scanner output to an LLM, I could build a privacy boundary between the security scanner and the model.

That made local AI a practical part of the security architecture rather than just a chatbot feature.

My Agent Session

I used AI-assisted development throughout the project, particularly for architecture exploration, debugging, testing, documentation, and implementation.

Prize Categories

Open-source AI / Local AI

Gemma 3 1B
Ollama
Local inference

Cybersecurity / Developer Tooling

Gitleaks
Secure finding sanitization
Repository security analysis

New To Community

Devansh Shukla — Mon, 05 Oct 2026 02:17:43 +0000

Hello Everyone!!!