This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass
What I Built
CyberEscape 🌿🔐 — Walk. Observe. Think like a defender.
What if learning cybersecurity didn't always mean sitting in front of a screen?
As someone learning and building in cybersecurity, I spend a lot of time working with virtual machines, terminals, networking tools, and security labs. These are essential for developing technical skills, but I wanted to explore another way of learning security concepts—one that connects everyday experiences with defensive thinking.
That's the idea behind CyberEscape, an AI-powered outdoor cybersecurity learning experience built for the Hacktoberfest 2026 Touch Grass challenge.
CyberEscape turns an ordinary walk into a safe, interactive cybersecurity mission. Instead of spending the entire experience looking at a screen, users receive a mission, put their phones away, observe their surroundings respectfully, and return to reflect on what they learned.
Here's how it works:
- Choose a mission: Select a duration, difficulty level, and cybersecurity topic.
- Generate with AI: Local AI creates an educational outdoor observation mission based on the selected preferences.
- Step outside: Read the instructions, put the phone away, and observe your surroundings without interacting with people or testing real security controls.
- Reflect: Return to the application and explain the cybersecurity principle, its potential impact, and a possible defensive improvement.
- Learn and progress: AI analyzes the reflection and provides educational feedback and defensive recommendations. Successful reflections earn 10 XP, while mission progress and history are stored in browser localStorage.
CyberEscape currently covers five topics:
- 🔐 Physical Security
- 🕵️ Privacy & Exposure
- 🎭 Social Engineering
- 🛡️ Defensive Thinking
- 🌐 Network Security Awareness
The idea isn't to replace hands-on cybersecurity labs. It's to complement them with a different kind of learning experience.
A traditional lab asks you to investigate a technical environment. CyberEscape encourages you to connect security principles with everyday situations, think about why they matter, and consider how a defender could reduce potential risks.
The screen starts the mission. The real world provides the observation. Reflection turns it into a learning experience.
Demo
The best way to understand CyberEscape is to see its actual workflow.
Application screenshots
The CyberEscape interface and mission progress.
Configuring a mission and receiving AI-generated instructions.
Submitting a reflection and reviewing AI-generated feedback.
🎬 Video demonstration:
The demo should show the actual application in action: selecting a mission, generating instructions, submitting a reflection, receiving feedback, and viewing the resulting progress.
These screenshots and the recording should come from the working MVP rather than mockups.
Code
🔗 GitHub repository: https://github.com/akashjha518/CyberScape
The repository contains the CyberEscape implementation and project documentation.
The project is built around a simple idea: combine an accessible web interface with local AI to create a learning experience that encourages people to spend less time looking at their screens.
How I Built It
CyberEscape is a full-stack application built with React, Python, FastAPI, and a locally running open-weight AI model.
Technology stack
| Technology | Role in CyberEscape |
|---|---|
| React | Builds the interactive user interface |
| Vite | Frontend development and build tooling |
| CSS | Application styling |
| Python | Backend implementation |
| FastAPI | API layer connecting the frontend to the application workflow |
| Pydantic | Validation of structured API data |
| Ollama | Runs the AI model locally |
Gemma 3 4B (gemma3:4b) |
Generates missions and analyzes reflections |
| Browser localStorage | Stores mission history and progress |
1. Mission generation
The learner selects the duration, difficulty, and cybersecurity topic. The frontend sends the relevant configuration through the backend workflow, where the local AI model generates an educational mission.
This is where AI makes the experience more adaptable than a fixed collection of lessons. Mission content can be generated according to the learner's selected preferences rather than requiring every learner to follow identical instructions.
2. Reflection and AI feedback
After completing the outdoor observation activity, the learner returns and submits a reflection.
The AI analyzes the explanation and provides feedback and defensive recommendations. This encourages the learner to move beyond simply noticing something and consider the underlying security principle, its potential impact, and possible mitigation.
The model provides educational guidance; its output should not be treated as a guarantee of security correctness.
3. Local AI with Ollama and Gemma 3 4B
CyberEscape uses Gemma 3 4B through Ollama for mission generation and reflection analysis.
I chose a local inference approach because it fits the project's educational and privacy-conscious goals. The application can run these AI interactions on the user's machine when the required model and runtime are installed and configured.
This also gives the project a practical way to experiment with open-weight AI without depending exclusively on a hosted commercial AI API.
4. Progress tracking
The application stores mission history and progress in browser localStorage. Successful reflections earn 10 XP, giving learners a lightweight way to track their progress without requiring a separate account system in this MVP.
The result is a connected learning loop: configure, generate, observe, reflect, receive feedback, and track progress.
Why Does Open Innovation Matter?
For CyberEscape, open innovation matters because it gives me greater flexibility in how the AI component is run, explored, and integrated into the learning experience.
Local inference and user control
Ollama and Gemma 3 4B allow the AI model to run locally on a user's machine, subject to hardware and setup requirements.
This is particularly relevant for an application that processes written reflections. In the intended local setup, mission and reflection processing can take place without sending that content to a hosted commercial AI inference API.
Local inference alone does not guarantee complete application privacy or offline operation. Other network requests and data-storage behavior still matter, and browser localStorage is not encrypted storage.
Experimentation with open-weight AI
Using an open-weight model makes it possible to investigate the AI workflow more directly, experiment with prompts, evaluate model behavior, and explore alternatives without building the entire experience around a single hosted model endpoint.
It also makes the AI component a useful learning opportunity in its own right. Developers can examine how the model fits into the application and understand how AI-generated content is incorporated into a practical workflow.
Why Gemma?
Gemma 3 4B is central to CyberEscape's two AI-powered activities: creating outdoor missions and analyzing learner reflections.
The model is not included merely to add an AI label to the project. Its role is to help personalize the mission experience and provide feedback after the learner returns.
Open-weight AI made this implementation approach possible while keeping local inference at the center of the design.
Safety by design
CyberEscape is an educational awareness tool, not a security-audit application.
Its missions do not ask learners to trespass, test real access controls, interact with strangers, collect personal information, photograph people, or scan and investigate real networks.
The outdoor activity is intended to involve respectful observation and defensive reasoning. Learners put their phones away during the mission and return to the application for reflection.
I wanted the experience to encourage security awareness without turning public spaces or other people into targets for experimentation.
My Agent Session
The official challenge template makes this section optional.
[Add your DevRelay agent session link here if you have a relevant session to share.]
If no suitable session has been saved, this section can be removed before publication. I don't want to present an unavailable session or invent an agent workflow that wasn't documented.
Prize Categories
Best Use of Gemma
CyberEscape uses Gemma 3 4B through Ollama for its AI-powered mission-generation and reflection-analysis workflows.
This category is relevant because Gemma is part of the actual implementation rather than a technology mentioned only in the project description.
Final Thoughts
CyberEscape started with a simple question: can cybersecurity learning encourage people to step away from their screens without losing the practical thinking that makes security education valuable?
My answer was to build a small, working MVP that combines AI-generated missions, outdoor observation, reflection, feedback, and progress tracking.
It doesn't replace technical labs, SIEM investigations, packet analysis, or hands-on defensive security practice. Instead, it explores another way to develop awareness and connect security principles with everyday experiences.
Building this project also reminded me that AI is most useful when it serves a clear purpose. In CyberEscape, AI generates the learning activity and helps interpret the reflection, while the learner remains responsible for observing, thinking, and explaining.
For me, that's the most interesting part of the project: technology helps create an experience that deliberately asks the user to put technology aside for a while.
CyberEscape 🌿🔐 — Walk. Observe. Think like a defender.
🔗 Explore the project: CyberEscape on GitHub
I'd love to hear how other developers are using open-source AI to make learning more practical, engaging, and connected to the world outside the screen.
Top comments (0)