Raj Patel — Intelligence Agency

OPERATIONAL SUMMARY

Subject specializes in cybersecurity, threat intelligence, and applied AI/ML for security. Expertise spans malware analysis and reverse engineering, software supply chain security, agentic AI systems for automated forensics, and cloud security architecture, with a growing focus on secure systems design and secure coding practices. Driven by a passion for automating workflows with AI, from agentic forensics pipelines to API-integrated automation systems. Currently embedded at Georgia Tech's COEUS/Astrolavos Lab, with additional field experience in cyber threat intelligence consulting (media & entertainment sector) and cyber risk consulting.

BADGES

Python

Javascript

AWS

Docker

Github

C++

Java

Bash

Kubernetes

Linux

Wireshark

Elasticsearch

PyTorch

NumPy

Pandas

n8n

ACADEMIC CREDENTIALS

TRAINING RECORD

M.S. COMPUTER ENGINEERING — Georgia Institute of Technology · January 2025–August 2026 (anticipated)

B.S. COMPUTER ENGINEERING — Georgia Institute of Technology · August 2020–December 2024

FIELD ASSIGNMENTS

ASSIGNMENT: FULL TIMEACTIVE

GRADUATE RESEARCH ASSISTANT — COEUS/ASTROLAVOS LAB, GEORGIA TECH

JANUARY 2026 – AUGUST 2026 · ATLANTA, GA

Built automated incident response pipeline using n8n with VirusTotal and sandbox telemetry integrations, cutting manual triage time by 90%.
Performed end-to-end security investigations correlating IDA Pro static disassembly with CAPE dynamic sandbox data, uncovering 25 hidden malicious behaviors across analyzed samples.
Developed an Agentic AI system generating structured analysis contexts and suppression-bypass scripts — 100% accuracy on known malware samples, 67% on held-out samples.

ASSIGNMENT: PART TIMECLOSED

CYBER RISK ANALYST (CONSULTING) — PRAMUKH TRANSPORT LTD.

AUGUST 2022 – JANUARY 2025 · KENYA · PART-TIME

Led a red team tabletop exercise and threat modeling engagement across 3 threat objectives: extortion, sabotage, and financial fraud.
Analyzed Water Ouroboros ransomware threat intelligence to identify 3 critical vulnerabilities in transport infrastructure, including exposed RDP services and weak credential management.

ASSIGNMENT: FULL TIMECLOSED

IT TECHNICIAN INTERN — AKSHAR AUTO SPARE HOUSE

JUNE 2020 – AUGUST 2020 · KENYA

Hardened 7 workstations through OS patching, vulnerability scanning, and password policy enforcement.

ACTIVE CASE FILES

FILE ID: 309

AGENTIC AI SYSTEM FOR MALWARE BEHAVIOR DISCOVERY

OBJECTIVE: Auto-locate anti-analysis triggers in malware and generate sandbox-environment patches to defeat them.

TECH STACK: Python, LLM-agent orchestration, static disassembly + dynamic API-call analysis

IMPACT: Uncovered 22 new malware behaviors on a class leaderboard, tied for the most of any submission, at $0.034 LLM cost.

[ ACCESS DOSSIER ]

FILE ID: 772

MALWARE REVERSE ENGINEERING

OBJECTIVE: Reproduce and extend MANTILLA malware analysis to improve static detection of packed binaries.

TECH STACK: Python, Ghidra, YARA, scikit-learn, pefile

IMPACT: Improved detection F1-score by 15% over baseline using enhanced feature engineering.

[ ACCESS DOSSIER ]

FILE ID: 233

CTI

OBJECTIVE: Aggregate and correlate open and commercial threat feeds into actionable intelligence dashboards.

TECH STACK: Python, Elasticsearch, MISP, STIX/TAXII

IMPACT: Reduced mean-time-to-detect by 35% across monitored assets.

[ ACCESS DOSSIER ]

FILE ID: 099

DEFENSIVE DOMAIN REGISTRATIONS

OBJECTIVE: Cost-effective, data-driven defensive domain-registration strategy against cybersquatting (team of 4).

TECH STACK: Python, pandas, NumPy, Matplotlib, tldextract, Jupyter

IMPACT: Strongest dispute-risk correlations found with company assets (r≈0.29) and OSINT-flagged activity (r≈0.35) across 27 Fortune 500 companies.

[ ACCESS DOSSIER ]

FILE ID: 277

CYBERSEC OF DRONES

OBJECTIVE: Detect and mitigate GPS spoofing and RF hijacking against UAV platforms in the field.

TECH STACK: Python, GNU Radio, SDR, Scapy

IMPACT: Detected spoofing within 1.2s and blocked 98% of hijack attempts in controlled tests.

[ ACCESS DOSSIER ]

FILE ID: 506

LLM SAFETY UNDER QUANTIZATION

OBJECTIVE: Quantify whether compressing an LLM for local deployment weakens its jailbreak guardrails (team of 4).

TECH STACK: Python, HuggingFace Transformers, BitsAndBytes, llama.cpp/GGUF, JailbreakBench

IMPACT: Quantization method mattered more than compression level; model choice mattered most of all.

[ ACCESS DOSSIER ]

FILE ID: 207

VLM ATTENTION

OBJECTIVE: Determine when VLMs genuinely use visual evidence vs. linguistic priors, across dialogue turns (team of 4).

TECH STACK: Python, BLIP-2, LLaVA-1.5, Qwen2.5-VL, attention rollout, COCO dataset

IMPACT: Qwen2.5-VL was the most visually grounded of the three models tested; BLIP-2 the least.

[ ACCESS DOSSIER ]

FILE ID: 711

ML-BASED IDS

OBJECTIVE: Remove backdoor poisoning from a deep-learning IDS via machine unlearning, cross-verified by a rule-based IDS (team of 3).

TECH STACK: Python, PyTorch, Suricata, CIC-IDS2017

IMPACT: Restored 100% detection of backdoored flows post-unlearning with clean accuracy unchanged at 99.27%.

[ ACCESS DOSSIER ]

FILE ID: 418

CYBER-PHYSICAL SYSTEMS

OBJECTIVE: Secure control-loop framework for resilient cyber-physical infrastructure under attack.

TECH STACK: C, RTOS, MATLAB/Simulink, MQTT

IMPACT: Sustained safe operation under simulated fault and false-data injection.

[ ACCESS DOSSIER ]

FILE ID: 655

DATA ANALYTICS - DRIVERS FOR CYBER CONFLICT

OBJECTIVE: Identify the real structural drivers of state-sponsored cyberattack targeting frequency (solo project).

TECH STACK: R, OLS regression, CFR Cyber Operations Tracker, V-Dem, World Bank Open Data

IMPACT: Internet penetration — not regime type — is the dominant predictor of targeting (p<0.01).

[ ACCESS DOSSIER ]

FILE ID: 824

SECURE CLOUD WEB APPLICATION

OBJECTIVE: Zero-trust cloud web application with hardened authentication and full audit logging.

TECH STACK: Node.js, React, AWS, Terraform, OAuth2

IMPACT: Passed external penetration test with zero critical findings.

[ ACCESS DOSSIER ]

FILE ID: 388

HANDHELD EMBEDDED DEVICE

OBJECTIVE: Design and build a handheld gaming device from raw microcontroller components (team of 4).

TECH STACK: C++, mbed LPC1768 SDK, RTOS, uLCD-144G2

IMPACT: Working maze/coin game across 3 difficulty levels with mutex-guarded multithreaded LCD rendering.

[ ACCESS DOSSIER ]

FILE ID: 512

ANDROID GAME DEVELOPMENT

OBJECTIVE: Apply core OOP design patterns to build a complete, testable Android game (team of 7).

TECH STACK: Java, Android SDK, JUnit 4, Mockito, Espresso

IMPACT: 2nd-most active contributor on the team (76 of 390 commits) — combat system, enemy Factory pattern, core screens.

[ ACCESS DOSSIER ]

SECURE CONTACT CHANNELS

Transmit through one of the verified contact vectors below:

To view the full history of field missions and technical training, download the primary personnel file:

DOWNLOAD RESUME.PDF