Home Research Areas

Focus Areas

Four interconnected domains covering the attack surface of modern AI systems — from model training to deployment, from evasion to exploitation.

AI Cyber Defense Secure AI Eng. Adversarial ML AI Red Teaming
01

AI-Powered Cyber Defense

// Overview

Modern security operations centers generate millions of log events, alerts, and telemetry data points daily — far beyond human analyst capacity. This research area explores how large language models, vector search, and RAG pipelines can be applied to automate threat detection, triage, and incident summarization at scale.

The focus is on practical, deployable tooling: systems that fit into real SOC workflows, augment analyst judgment, and surface the high-fidelity signals buried in noise.

// Why It Matters

The Scale Problem

The average enterprise SIEM generates 10,000+ alerts per day. Analysts can only triage a fraction. AI-assisted defense isn't a nice-to-have — it's a structural necessity. Companies that get this right will have a decisive security advantage.

// My Contributions

Designing RAG-based log ingestion pipelines that surface semantic anomalies using embedding similarity
Building LLM-powered alert summarization and narrative generation for Tier-1 SOC workflows
Evaluating prompt injection risks in security AI tools — both theoretical and practical
Creating evaluation benchmarks for AI threat detection accuracy vs. traditional SIEM rules

// Key Technologies

PythonLangChainChromaDB OpenAI APIElastic SIEMSplunk LangGraphFastAPI

// Related Projects

AI Log Analyzer → AI SOC Assistant →

Research Status

Active
Overall Progress 35%

Key Concepts

Retrieval-Augmented Generation (RAG)
Semantic similarity search
Alert triage automation
LLM-assisted DFIR

Target Roles

AI Security Engineer
Detection Engineer (AI)
Security Data Scientist
02

Secure AI Engineering

// Overview

Shipping a model to production without security controls is the AI equivalent of deploying an unauthenticated web server. This research area is about building the security primitives that should be standard in every ML pipeline: model scanning, dependency pinning, secrets management, provenance tracking, and supply-chain integrity.

The goal is to produce a reusable, production-grade Secure MLOps Template that any team can fork and deploy — engineering security culture into the workflow itself.

// Why It Matters

The Supply Chain Gap

Model weights downloaded from public hubs have no mandatory verification. A poisoned model checkpoint can persist undetected through fine-tuning. Most ML teams have no SBOM for their models. This is a solved problem in software security — it just hasn't been ported to AI yet.

// My Contributions

Designing a Secure MLOps Template with GitHub Actions CI/CD, model scanning, and dependency audits
Implementing model provenance tracking using cryptographic hashing and signed manifests
Integrating secrets management (Vault / SOPS) into ML training and inference pipelines
Building an AI threat model framework inspired by MITRE ATLAS and STRIDE
Documenting best practices for containerized model serving with runtime security controls

// Key Technologies

GitHub ActionsDockerHashiCorp Vault SigstoreSOPSTrivy BanditSafety CLI

// Related Projects

Secure MLOps Template → AI Threat Modeling Framework →

Research Status

Planned
Overall Progress 10%

Key Concepts

Model supply-chain security
ML SBOM & provenance
Secure MLOps pipelines
Container runtime security

Target Roles

MLSecOps Engineer
AI Platform Security
ML Infrastructure Security
03

Adversarial Machine Learning

// Overview

Adversarial ML sits at the intersection of security and deep learning: small, carefully crafted perturbations to inputs can fool state-of-the-art neural networks with high confidence. This research area covers the full attack surface — white-box attacks like FGSM, PGD, and C&W, black-box transfer attacks, and the defenses that actually hold up under pressure.

The goal is not just to implement attacks but to understand the geometry of feature spaces, evaluate robustness certifications, and produce reproducible experiments that cut through the hype in the literature.

// Why It Matters

High-Stakes Evasion

Adversarial examples have been demonstrated against autonomous vehicle perception, medical imaging classifiers, malware detectors, and facial recognition systems. When AI makes safety-critical decisions, adversarial robustness isn't academic — it's a product requirement.

// My Contributions

Implementing FGSM, PGD, Carlini-Wagner, DeepFool, and Universal Adversarial Perturbations in PyTorch
Reproducing 3–5 landmark adversarial ML papers with annotated code and commentary
Building an interactive attack playground with visualizations of decision boundaries and perturbation norms
Evaluating adversarial training, certified defenses, and randomised smoothing on standard benchmarks

// Key Technologies

PyTorchART (IBM)Foolbox CIFAR-10ImageNetCleverHans W&B SweepsMatplotlib

// Related Projects

Adversarial Attack Playground → Paper Reproductions →

Research Status

Planned
Overall Progress 5%

Attack Families

FGSM / I-FGSM
PGD (Madry et al.)
Carlini & Wagner (C&W)
DeepFool
Universal Perturbations

Target Roles

Adversarial ML Researcher
AI Safety Engineer
ML Security Researcher
04

AI Red Teaming & Evaluation

// Overview

AI red teaming is the systematic adversarial testing of AI systems — going beyond functional QA to probe safety boundaries, alignment failures, and attack surfaces that only emerge under adversarial pressure. This research area builds the tooling and methodology to make AI red teaming rigorous, repeatable, and scalable.

The north star is the LLM Red Teaming Evaluation Suite: an automated framework that covers jailbreaks, safety bypasses, toxicity elicitation, PII extraction, and prompt leakage — producing structured reports that inform safety alignment work.

// Why It Matters

Safety Cannot Be Asserted — It Must Be Demonstrated

RLHF and Constitutional AI reduce but do not eliminate harmful outputs. Red teaming uncovers the cases that slip through safety training. Organizations like OpenAI, Anthropic, and Google DeepMind now run dedicated red teams before every major model release — and hire accordingly.

// My Contributions

Building an automated jailbreak testing suite with structured attack taxonomies and success metrics
Developing toxicity and PII extraction evaluation harnesses using standardized LLM judge frameworks
Documenting attack patterns against GPT-4, Claude, and open-weight models with reproducible prompts
Mapping attack patterns to MITRE ATLAS tactics, techniques, and sub-techniques
Creating a structured red-team report template that aligns with NIST AI RMF disclosure expectations

// Key Technologies

PythonPyRITGarak LangSmithMITRE ATLASOpenAI Evals NIST AI RMFLLM-as-Judge

// Related Projects

LLM Red Teaming Suite → Prompt Injection Toolkit →

Research Status

Planned
Overall Progress 8%

Attack Vectors

Jailbreak prompts
Safety bypass patterns
Toxicity elicitation
PII & data extraction
Prompt leakage

Target Roles

AI Red Team Lead
Trust & Safety Engineer
AI Safety Researcher