Research portfolio / security systems / explanatory AI

I build security tools people can trust at a glance.

Cybersecurity, AI safety, and robotics projects presented with a clear story, visible proof, and one memorable visual signature.

View Projects Read Blog Resume →

Interpretable

Security AI and model outputs that can be explained, traced, and audited.

Operational

Tools designed to plug into real workflows, not just read well in a demo.

Open

Built in public with concrete repos, blog posts, and reusable components.

Drag the orbs · Click to pause · Zero DOM reads

What sticks

Clear intent, real proof, and a specific point of view

Interpretable systems

I focus on security tools that explain their own decisions instead of hiding behind a black box.

Actionable workflows

Every project is built to fit into a real process: scanning, verification, alerting, or review.

Visible output

The work is public, documented, and grounded in specific outcomes rather than vague buzzwords.

Featured work

Selected projects with the clearest story

Proxy / verification

VeriAlign

Turns LLM API traffic into verified responses with claim grounding, contradiction checks, and confidence metadata.

LLM APIs Source grounding Response verification

Audit / plugins

Community AI Audit

A plugin-driven framework for scanning AI behavior across providers and sending findings to SIEM tools.

Multi-provider Plugin architecture SIEM output

Dashboard / defense

ORYTH

An email security dashboard for infrastructure discovery, SPF/DKIM/DMARC checks, and spoofing risk analysis.

CENSYS Threat intel Email posture

Research / explainability

Prompt Injection Visualizer

A browser-native educational platform that classifies prompt injection attacks using Transformers.js, Tokenizers.js, and D3.js — all running locally via WebGPU.

Transformers.js WebGPU D3.js ONNX Runtime

My focus

What I try to make obvious in every build

🤖

Robotics

Exploring embodied intelligence, real-time control systems, and the intersection of hardware and software. From sensor fusion to autonomous navigation — building machines that perceive and act in the physical world.

🧠

Artificial Intelligence

Building interpretable Small Language Models (SLMs) for security-critical applications. Researching agentic frameworks, GRPO training, and model interpretability at the edge.

🔐

Cybersecurity

Blue-team defense with full transparency. From vulnerability research to autonomous security agents — every decision traced, every alert explained. Trust no input. Verify all output.

All projects

Open source tools and research

Community AI Audit

● Active

Plugin-driven framework for auditing AI model behavior across providers and deployment targets.

Python AI Safety Plugin Architecture Multi-Provider Splunk Elastic Datadog Sentinel

GitHub →

ORYTH

● Active

Email security dashboard powered by CENSYS — monitor, analyze, and defend your email infrastructure.

Python CENSYS Email Security Dashboard Threat Intelligence

GitHub →

Prompt Injection Visualizer

● Active

A browser-native educational platform that classifies, visualizes, and explains prompt injection attacks using Transformers.js, Tokenizers.js, and D3.js — al...

React 19 TypeScript Transformers.js Tokenizers.js ONNX Runtime Web WebGPU D3.js Monaco Editor TailwindCSS Zustand Framer Motion

Launch App → Source →

VeriAlign

● Active

A reverse proxy that intercepts LLM API calls and augments responses with verification metadata.

Python Reverse Proxy LLM APIs Claim Extraction Source Grounding Contradiction Detection

GitHub →

VulnGPT MCP Server

◉ Completed

Vulnerability scanning with Model Context Protocol for Puch AI Hackathon.

Python MCP Vulnerability Scanning Puch AI Hackathon

GitHub →

Read the Blog

Cybersecurity research, AI/ML deep dives, and robotics experiments — written in the open.

Visit Blog →