For employers

Hire this AI Trainer

Sign in or create an account to invite AI Trainers to your job.

LoginCreate account
Invite to Job
Mel De Silva

Mel De Silva

CTO & Co-Founder - Cybersecurity & AI

SRI_LANKA flag
Colombo, Sri Lanka
$40.00/hrExpertLabelboxLabelimgProdigy

Key Skills

Software

LabelboxLabelbox
LabelImgLabelImg
ProdigyProdigy
RoboflowRoboflow
Scale AIScale AI
Other

Top Subject Matter

No subject matter listed

Top Data Types

Computer Code ProgrammingComputer Code Programming
DocumentDocument
ImageImage
TextText
VideoVideo

Top Label Types

Bounding Box
Classification
Entity Ner Classification
Red Teaming
RLHF
Routing
Segmentation
Text Generation
Text Summarization
Tracking

Freelancer Overview

I am a deep learning researcher and cybersecurity practitioner with hands-on experience spanning computer vision, transfer learning, and offensive AI security. My work focuses on building high-performance AI systems while systematically evaluating and hardening them against real-world cyber threats, adversarial manipulation, and misuse scenarios. I have designed and implemented deep learning solutions for document enhancement, classification, and detection, supported by robust data pipelines for simulation and sensor data collection in autonomous and safety-critical environments. Alongside this, I actively conduct AI red-teaming and security assessments, testing AI and LLM-powered systems for vulnerabilities such as prompt injection, jailbreak techniques, data poisoning, model extraction, model inversion, insecure fine-tuning, and abuse of agent-based workflows. My cybersecurity experience includes assessing AI-integrated web applications, APIs, and cloud-hosted platforms, where I evaluate attack surfaces introduced by LLM integrations, retrieval-augmented generation (RAG), and autonomous agents. I have performed controlled jailbreak testing to identify weaknesses in alignment, safety guardrails, and prompt handling logic, helping organizations strengthen policy enforcement, monitoring, and defensive controls before production deployment. Technically, I work extensively with TensorFlow, Python, and backend technologies such as Node.js, integrating AI models into secure, production-grade systems. I have also led hands-on courses and labs in machine learning, deep learning, reinforcement learning, and secure AI deployment, translating advanced research concepts into operationally safe implementations. A key pillar of my work is data and model trust—ensuring that datasets, annotations, and training pipelines are resilient against tampering, leakage, and bias, while maintaining high model accuracy and reliability. I am particularly passionate about applying adversarial testing, threat modeling, and red-team methodologies to ensure AI systems are not only intelligent, but robust, explainable, and secure by design.

ExpertEnglishGermanSinhalese

Labeling Experience

Adversarial Data Labeling & Prompt Annotation for AI Red Teaming and Jailbreak Testing

OtherTextText GenerationFine Tuning
Conducted structured data labeling and annotation for AI red teaming and controlled jailbreak testing engagements. Work involved labeling prompt-response pairs to identify security risks such as prompt injection, policy bypass attempts, unsafe completions, and misuse scenarios in LLM-powered systems. Developed annotation taxonomies for risk severity, exploitability, and mitigation priority, supporting security assessments, model hardening, and guardrail validation. Labeled datasets were used to improve detection logic, safety evaluations, and governance controls for enterprise AI deployments.

Conducted structured data labeling and annotation for AI red teaming and controlled jailbreak testing engagements. Work involved labeling prompt-response pairs to identify security risks such as prompt injection, policy bypass attempts, unsafe completions, and misuse scenarios in LLM-powered systems. Developed annotation taxonomies for risk severity, exploitability, and mitigation priority, supporting security assessments, model hardening, and guardrail validation. Labeled datasets were used to improve detection logic, safety evaluations, and governance controls for enterprise AI deployments.

2025

Research-Grade Data Labeling & Annotation for Computer Vision and Transfer Learning

OtherImageBounding BoxSegmentation
Performed research-grade data labeling and annotation as part of university-led deep learning and computer vision research. Work included annotation of image, video, and sensor datasets for object detection, semantic segmentation, and classification tasks supporting transfer learning and domain adaptation experiments. Responsibilities included dataset curation, annotation guideline development, quality assurance, and validation to ensure consistency and statistical reliability across training and evaluation datasets. Data was used in safety-critical and research environments such as autonomous systems, robotics, and document analysis.

Performed research-grade data labeling and annotation as part of university-led deep learning and computer vision research. Work included annotation of image, video, and sensor datasets for object detection, semantic segmentation, and classification tasks supporting transfer learning and domain adaptation experiments. Responsibilities included dataset curation, annotation guideline development, quality assurance, and validation to ensure consistency and statistical reliability across training and evaluation datasets. Data was used in safety-critical and research environments such as autonomous systems, robotics, and document analysis.

2019 - 2022

Education

C

Carinthia University of Applied Sciences

Master of Science, Systems Design

Master of Science
2013 - 2015
K

Kingston University

Bachelor of Engineering, Aerospace Engineering Design

Bachelor of Engineering
2009 - 2012

Work History

R

Red Threat Cyber Security

Chief Technology Officer

Ganemulla
2022 - Present
R

Red Threat Cyber Security

Co-Founder & Chief Technology Officer

Gampaha, Sri Lanka
2022 - Present