📌 Introduction

This article chronicles my journey from conducting academic research on Integrated Sensing and Communications (ISAC) security for my MSc dissertation at Kingston University, London, to building a production-ready Multi-Agent GenAI System using Azure OpenAI.

The project bridges two worlds: traditional cybersecurity research (encryption, vulnerability assessment, network security) and modern Generative AI (multi-agent systems, function calling, LLM orchestration).

🎓 My MSc Dissertation: Secure Transmission and Security Assessment of ISAC Applications

The Problem

My dissertation, completed in September 2024 at Kingston University's Faculty of Science, Engineering and Computing, focused on a critical question:

How can we securely transmit health data in Integrated Sensing and Communications (ISAC) applications while ensuring confidentiality, integrity, and availability?

What is ISAC?

ISAC (Integrated Sensing and Communications) is an emerging technology that merges sensing (collecting data from sensors like ECG, accelerometer, gyroscope) and communication (transmitting that data over 5G/WiFi) into a single integrated system.

In traditional systems, sensing and communication are separate. ISAC does both simultaneously using the same hardware and frequency. This creates unique security challenges.

Research Objectives

Technology Stack

• MATLAB for health data simulation and visualization

• Python for server-side encryption (AES-256, RSA-2048)

• TCP/IP sockets for client-server communication

• Kali Linux for penetration testing

• Metasploit framework for exploitation (MS08-067)

Vulnerability Assessment Results

Key Findings

Research Gaps Identified

1. Limited ISAC-specific security research - Most work focuses on either sensing OR communication, not both

2. No multi-agent approach for ISAC security - Traditional tools are single-purpose

3. Lack of GenAI integration in security assessment - LLMs not used for threat analysis

4. Manual vulnerability assessment - No automation or intelligent analysis

These gaps directly inspired my GenAI project.

🤖 The Evolution: From Traditional Security to GenAI Multi-Agent Systems

Why GenAI for ISAC Security?

Traditional security tools (Nmap, Metasploit) identify vulnerabilities but don't analyze or recommend solutions. They tell you what is wrong, not what to do about it.

GenAI agents can:

• Understand the security context

• Analyze vulnerabilities across domains

• Generate actionable recommendations

• Integrate sensing and communication analysis (the "I" in ISAC)

The 4 Specialized Agents

Sequential Pipeline (Output Chaining)

INPUT → AGENT 1 → AGENT 2 → AGENT 3 → AGENT 4 → OUTPUT

Each agent's output becomes the next agent's input.

Function Calling Tools

# Tool 1: Signal Anomaly Detection
def detect_signal_anomaly(snr_db, rssi_dbm, protocol):
    if snr_db < 5:
        return {"anomaly_type": "JAMMING_DETECTED", "risk_level": "HIGH"}
    return {"anomaly_type": "NORMAL", "risk_level": "LOW"}

# Tool 2: Threat Score Calculator
def calculate_threat_score(threat_type, encryption, attack_success):
    base_score = {"Spoofing": 80, "Jamming": 70, "Tampering": 85}
    score = base_score.get(threat_type, 50)
    return {"threat_score": score, "risk_level": "CRITICAL" if score > 70 else "HIGH"}

📊 Comparison: Traditional vs GenAI Approach

💡 What I Learned

From My MSc Dissertation:

1. Encryption is necessary but not sufficient - Perfect encryption doesn't protect against system-level exploits

2. Cross-platform compatibility is challenging - MATLAB to Python encryption required careful byte handling

3. Vulnerability assessment is time-consuming - Manual scanning requires expertise and time

4. Legacy systems are the weakest link - Outdated systems compromise security

From Building the GenAI System:

1. AI agents can automate security analysis - What took hours now takes 30 seconds

2. Sequential pipelines enable complex reasoning - Each agent builds on the previous

3. Function calling bridges LLMs and quantitative analysis - Agents calculate exact threat scores

4. ISAC requires integrated security thinking - Sensing and communication vulnerabilities cascade

5. User-friendly interfaces matter - Streamlit UI makes security analysis accessible

6. Safety guardrails are essential - max_iterations=3 prevents infinite loops

📈 Results

Dissertation Results

GenAI System Results

Key Findings

Recommendations Generated

🚀 Future Directions

🔗 Resources

• GitHub Repository: https://github.com/sravsperum1/isac-health-security

• Dataset: https://www.kaggle.com/datasets/uciml/human-activity-recognition-with-smartphones

• Azure OpenAI Docs: https://learn.microsoft.com/en-us/azure/ai-services/openai/

• Streamlit Docs: https://docs.streamlit.io/

Built with: Python, Azure OpenAI GPT-4o, Streamlit, Pandas, Matplotlib, Seaborn