Looking for a job

TONGA MARC-DONALD

@tongamarc-donald

AI red-team engineer behind PromptShield, whose tests surfaced an encoding paradox across 11 providers; also building redforge and AI-IDS.

Cameroon

Message

With PromptShield, I engineered adversarial prompts to test LLMs across providers and uncovered a novel “encoding paradox” vulnerability. The evaluation measured a 66.7% encoding-attack success rate on Nemotron Ultra 550B versus 40.0% on Llama 3.3 70B.

I also built redforge, an automated LLM red-teaming toolkit, and AI-IDS, a network intrusion detection project. Through self-employed client work, I shipped production web applications with security protections, and I’ve contributed open-source cybersecurity tools for port visibility, vulnerability scanning, and network monitoring.

Experience

Work history, roles, and key accomplishments

Current

AI Red Team Tooling Engineer

Current

redforge

Jan 2026 - Present (9 months)

Built an automated LLM red-teaming toolkit that runs 10 attack families against any model endpoint, as measured by ~1,800 lines of code across 31 modules with four pytest suites, by creating a provider-agnostic adapter layer over async httpx with auto-discovery attack registration.

LLM Security Red Teaming Python Async Programming Httpx Pytest Adapter Translator Pattern Automation

Current

AI Consultant & Full-Stack Developer

Current

Self-Employed

Jan 2018 - Present (8 years 9 months)

Shipped 10+ production web applications with zero reported security incidents across seven years of bilingual client engagements by integrating input validation, CSRF protection, secure authentication, and API hardening into every build.

Fullstack Development Web Applications Security Input Validation CSRF Authentication API Hardening JavaScript React NodeJS Python Django Flask

AI Red Team Engineer & LLM Evaluation Researcher

PromptShield

Jan 2025 - Jan 2026 (1 year)

Uncovered a novel 'encoding paradox' vulnerability in frontier LLMs, as measured by a 66.7% encoding-attack success rate on Nemotron Ultra 550B versus 40.0% on Llama 3.3 70B, by engineering 55 adversarial prompts across eight injection categories and evaluating 67 model configurations across 11 providers.

LLM Security Prompt Injection JailBreak Obfuscation Prompt Extraction Multi Turn Escalation AI Evaluation Red Team Rubric Based Scoring Preference Ranking Safety Labeling LLM As A Judge Python PyTorch Hugging Face Transformers

Machine Learning Security Engineer

AI-IDS

Jan 2024 - Jan 2025 (1 year)

Achieved near-perfect F1 classification of 14 network attack classes, as measured by 2.8M+ labeled flows processed with 180% expanded attack-class coverage, by training a two-stage Random Forest pipeline on CIC-IDS2017 and building a real-time flow assembly module in Python and Scapy.

Machine Learning Random Forest Python Scapy Network Security Intrusion Detection ICIDS V Data Processing

Cybersecurity Tool Developer

Automatic-Porter, Vulnera & NMS

Jan 2024 - Jan 2025 (1 year)

Delivered three open-source security tools for port visibility, vulnerability scanning, and network anomaly monitoring, as measured by 15 maintained public repositories plus GitHub Pull Shark and YOLO achievements, by designing each tool around real-time detection and alerting workflows.

Cybersecurity Open Source Port Scanning Vulnerability Scanning Network Monitoring Real Time Detection Alerting GitHub

Education

Degrees, certifications, and relevant coursework

ICT University

Bachelor of Science, Cybersecurity

Pursuing a B.Sc. in Cybersecurity with expected graduation in 2026. Coursework includes Machine Learning, Network Security, Cryptography & Cryptanalysis, Information Systems Security, Linux Administration, and Digital Forensics.