CyberGuard
CyberGuard brings AI-powered phishing, QR, voice, media, and malware detection together in one explainable cybersecurity platform.
- Video of it working
- Problem, approach and result explained
- Public source code
- Answered the full brief· 13 of 13
- Expert reviewed· 6 reviews
—
Core threat-analysis areas integrated
—
Application platforms: Web, Windows Desktop, and Android
—
Static malware scanning engines: ClamAV and YARA
Walkthrough
Overview
The problem
The sophistication of cybersecurity threats continues to escalate, creating a paramount challenge: the growing difficulty in discerning authentic activities from malicious intrusions.
Malicious digital artifacts often masquerade with deceptive precision: a fraudulent hyperlink may mimic the appearance of a trusted institution, while a QR code can covertly reroute users to sophisticated phishing landscapes. Furthermore, artificial intelligence-driven voice synthesis enables the creation of audio recordings that replicate the timbre and cadence of loved ones or corporate leaders with startling authenticity.This digital deception extends beyond audio and text; photographs, video footage, and executable software files are increasingly susceptible to manipulation or engineered specifically to subvert user trust.
Ordinary internet users, students, employees, businesses, and organizations are all susceptible to such attacks. The repercussions may encompass the compromise of passwords, financial fraud, identity theft, and data breaches.
The primary objective was to resolve the fragmentation inherent in current security practices, where users are compelled to rely on disparate tools to detect various threat vectors. Furthermore, many of these instruments merely issue alerts without elucidating the underlying rationale, thereby offering little context or actionable insight.
Our objective was to develop a more pragmatic solution: an integrated security ecosystem capable of analyzing diverse categories of suspicious digital assets, discerning potential threats, and elucidating the evidentiary basis for its conclusions.
This concept constitutes the foundational premise of CyberGuard.
The approach
To address this challenge, we engineered CyberGuard as a multi-layered cybersecurity and threat intelligence platform, thereby circumventing the limitations inherent in relying on any singular detection methodology.
The architecture is anchored by a Python-based FastAPI backend that functions as the central processing engine, interfacing with a web-based security dashboard while simultaneously supporting dedicated desktop and Android components.
Upon the submission of potentially suspect material, the backend system classifies its nature and directs it to the designated analysis engine.
To detect suspicious URLs, CyberGuard analyzes anomalous domain structures, deceptive character substitutions, Punycode encoding, URL entropy metrics, and redirection behaviors. Furthermore, the system correlates this data with external threat intelligence feeds, including those from VirusTotal and URLhaus.
Regarding QR codes, the system initially deciphers the embedded data and subsequently evaluates the generated URL through the identical threat-detection methodology.
In the realm of audio, imagery, and video verification, we deploy artificial intelligence models in tandem with established forensic methodologies. Specifically, image forensics leverages Error Level Analysis and frequency decomposition to detect subtle signatures indicative of manipulation.Concurrently, voice authentication scrutinizes acoustic characteristics and supplementary speaker identity markers to ascertain authenticity.
The solution further integrates a comprehensive malware analysis module, synthesizing file structure inspection, ClamAV scanning, YARA rule matching, and entropy analysis. Crucially, the examination of suspicious files is conducted strictly in a static manner, thereby precluding any execution on the user's operating system.
The result
The outcome is CyberGuard, an integrated cybersecurity prototype that consolidates diverse threat-analysis capabilities within a unified platform.
Rather than toggling among disparate utilities, users can leverage a unified dashboard to execute a range of security assessments.
For instance, a user may submit a suspect website URL, thereby triggering an analysis grounded in the domain's specific attributes, its redirect mechanisms, and existing threat intelligence.
In parallel, a QR code may be scanned to retrieve its concealed destination prior to any assessment of said destination. Furthermore, audio files, images, video recordings, and potentially malicious documents can be transmitted to their dedicated analytical engines for scrutiny.
Rather than merely assigning a binary SAFE or UNSAFE designation, the system delivers comprehensive risk classifications alongside the evidentiary findings that underpin them. This approach empowers both end-users and security analysts to discern the precise rationale behind any flagging decision.
Reflection
Throughout the course of this project, we have come to recognize that constructing a cybersecurity platform extends far beyond the mere integration of artificial intelligence models or the incorporation of external security APIs.
A primary insight we have garnered is that no solitary detection methodology can be deemed universally reliable across all operational contexts.
Discrepancies frequently arise when distinct forensic methodologies or computational models yield contradictory outcomes; consequently, the rigorous correlation of evidence and the transparent execution of risk assessments become paramount.
Submitted to CYBERGUARD: Trust Nothing, Explain Everything
- Your idea in one line
CyberGuard is an AI-powered cybersecurity operations platform that detects phishing, malicious URLs, suspicious images, voice deepfakes, and technical cyber threats, then correlates threat intelligence, calculates risk scores, and supports security teams with centralized monitoring, incident analysis, and policy-based response.
- What will be working by the finale?
By the Grand Finale, we aim to demonstrate a working CyberGuard prototype with these features: 1. Threat detection for suspicious URLs, phishing, files, images, and voice recordings. 2. Threat intelligence using VirusTotal and URLhaus. 3. Risk scoring, incident tracking, and a centralized SOC dashboard. 4. Image and deepfake analysis using image patterns, edges, and available metadata. 5. Security administration with access controls, policies, audit records, and supported response actions. 6. Email alerts using a free email service, with delivery tested. 7. Testing of API connections, database storage, detection workflows, error handling, and dashboard responsiveness. We will clearly show which features are working and tested and which are still under development. Advanced features such as device blocking and DDoS protection will be presented only if they are implemented and tested.
- Tech stack and hardware you plan to use
Programming languages: Python, JavaScript, HTML5, and CSS3. Backend: Python 3.12, Uvicorn, REST APIs, and WebSockets. Frontend: HTML, CSS, JavaScript, and a responsive SOC dashboard with visualizations. AI and machine learning: PyTorch, Wav2Vec2, ECAPA-TDNN, MFCC, and log-Mel spectrograms for audio analysis. Image analysis uses FFT, edge detection, noise analysis, and available metadata. A validated image classification model is a future improvement. Threat intelligence: VirusTotal API v3 and URLhaus for checking supported URLs, domains, IP addresses, and file hashes. Database: SQLite for storing incident records, application data, and audit logs. Security and administration: JWT authentication, role-based access control, risk scoring, security policies, rate limiting, and email notifications. Security frameworks: MITRE ATT&CK and NIST incident-response guidance, where applicable. External services: VirusTotal, URLhaus, and an email service such as Brevo. Hardware: A standard laptop or PC, a web browser, and an internet connection. An Android device is useful for testing supported monitoring features. A dedicated GPU is not required for basic operation. Additional requirements: API keys, verified email configuration, test images and audio files, safe and malicious test URLs, and a controlled testing environment. The project mainly uses open-source tools and free service plans to reduce costs. Deployment is optional.
- What have you built or tested so far?
We have developed CyberGuard, a multi-modal cybersecurity prototype with a Python FastAPI backend, web-based Security Operations Center (SOC) dashboard, and native Android companion application. The platform combines voice-deepfake analysis using PyTorch, Wav2Vec2, and ECAPA-TDNN with phishing and URL analysis, QR-code forensics, image-forensic heuristics, and threat intelligence through VirusTotal and URLhaus. The system includes real-time audio telemetry using WebSockets, SQLite-based persistence, risk scoring, incident management, JWT authentication, role-based access control, administrative policy controls, and model-review and lifecycle workflows. We have also tested selected backend APIs, neural inference, threat-scoring logic, threat-intelligence integrations, database persistence, WebSocket connectivity, and security-enforcement paths. Current work focuses on improving image and deepfake classification accuracy, validating complete detection-to-response pipelines, making forensic analysis evidence-grounded, and completing end-to-end email notification and regression testing. Image analysis currently uses FFT, Laplacian, and metadata-based forensic techniques, so further validation is needed before claiming reliable general-purpose AI-image detection. Our goal is to demonstrate a functional prototype while clearly identifying limitations and distinguishing verified results from ongoing development.
- Pitch deck (PDF)
- Pitch deck (PDF)PDF
- GitHub repository
- github.com/Arthur-2407/CyberGuard
- What's working now?
1. URL and Phishing Detection We will start by testing a suspicious URL and showing how CyberGuard checks the website, identifies possible phishing signs, and assesses its risk using available threat intelligence. 2. QR Code Threat Detection Next, we will upload a QR code to show how CyberGuard extracts its hidden link and checks whether the destination could be harmful or unsafe. 3. AI Voice Fraud Detection We will test genuine and AI-generated voice recordings to demonstrate how CyberGuard analyses voice characteristics and looks for signs of voice cloning or manipulation. 4. Malware Analysis We will upload a sample file for security inspection. CyberGuard will check its file structure, generate its SHA-256 hash, analyse its entropy, and use tools such as ClamAV and YARA to identify possible threats without executing the file. 5. Explainable Threat Intelligence Finally, we will demonstrate how CyberGuard brings together security findings to assess potential threats, explain why something may be suspicious or malicious, and suggest suitable actions. Our aim is to show that CyberGuard not only detects potential threats but also helps users understand them and take informed security decisions.
- GitHub repository
- github.com/Arthur-2407/CyberGuard
- Deck or report (PDF)
- Deck or report (PDF)PDF
AI tools used
.