Ask the local mentor
The FastAPI application connects to Ollama for local model inference and guided hints.
CyberGuide-AI
A better question. A deeper understanding.
A CTF learning platform that brings guided practice and a local AI mentor together.
QuestionPracticeUnderstand
ARCHITECTURE STUDY · ILLUSTRATIVE
A hint is a step, not the finish line.
Illustrative learning journey assembled from documented features. This is not a screenshot or a generated chat transcript.
Work through the challenge
The learner investigates a CTF challenge and submits a flag. No fabricated answer or success rate is shown here.
Keep the learning trail
Persistent progress tracking and writeups support revisiting the work; WebSocket updates serve the live leaderboard.
AI supports the learning process. The platform still separates guidance, challenge submissions and recorded progress.
Read the source ↗01 / The question
Security learning needs room to investigate, not just an answer. How can a local assistant support a learner while challenges and progress remain part of one experience?
02 / Project contribution
A school project combining CTF challenges and an Ollama-backed mentor with a FastAPI backend. The repository documents frontend, backend and AI roles without a verified individual allocation; this study describes the project implementation.
03 / Inside the implementation01 02 03
Practice in context
Accounts, challenge tracking, flag submission and writeups keep the learning journey connected. SQLite persists user and challenge data.
Ask a local mentor
Ollama provides local model inference. Prompt and walkthrough files guide the assistant; the model is configurable.
Connect the platform
JWT authentication and admin checks sit alongside competition events and a WebSocket leaderboard. Docker Compose brings the application and local model service together.
04 / Scope & lessons
A school learning platform with no verified production deployment. The first registered account becomes administrator; chat sessions use in-process memory. README model defaults differ between sections, so no fixed model is claimed here. Local inference quality and resource needs depend on the chosen model.