Projects

A selective map of shipped systems, research, and experiments. The strongest signals are production impact, applied AI judgment, and moving quickly without hiding risk.

All projects shown

AI & ML Research

HERA

Hierarchical Ensemble Reasoning Agent, Thesis A 3-layer Google ADK multi-agent system to mitigate correlated LLM errors with ensemble voting. Integrates hybrid Code World Model (AST + neural) and GEPA gradient-free prompt optimizer across 14 tools. Benchmarked 32B model: 16.0% SWE-bench Verified Mini, 24.5% LiveCodeBench, 71.0% CruxEval.

View Live Demo

ControlKeel

Governance control plane for AI coding agents. I built deterministic policy validation, review gates, proof capture, and routing across agent hosts. Agents can move quickly, but risky execution should remain inspectable and explicitly governed.

View Project · View GitHub

⚓ ControlKeel Governed agent execution

Society of Kawaii

Multi-Agent Debate Arena with Prediction Markets. A multi-agent debate arena where AI agents with opposing worldviews battle for truth. Inspired by Minsky's "Society of Mind," featuring real-time Google Search grounding, Elo-based prediction markets, and evolutionary mutations.

View Live Demo · View Write-up

👾 Society of Kawaii societyofkawaii.com

OS-Ghost: The Living Desktop

Autonomous, multi-modal desktop companion. A screen-aware desktop agent built with Rust and Tauri v2 that sees your screen using Gemini Vision, orchestrates workflows into an interactive mystery game, and features a multi-agent system with Orchestrator, Critic, Planner, and Guardrail agents.

View Devpost · View GitHub

ScreenSafe

Privacy Protection and on-device PII detection. ScreenSafe is an on-device PII detection application built with React Native and Cactus. It demonstrates how to use local LLMs and vision models to protect user privacy by detecting sensitive information in images.

View GitHub · View Blog

Memory Palace MCP

A memory enhancement tool leveraging LLM technology. An MCP server that turns source material into spatial memory-palace scenes, combining a classic recall technique with an interface AI agents can call directly.

View Live Demo · View GitHub

Climate Change, Hurricanes, and Property Loss

A Machine Learning Approach to Studying Infrastructure Sustainability This study examines how hurricane intensity and persistence influence infrastructure loss. Using data from the NOAA Storm Events Database, we develop machine learning models to predict property losses based on storm characteristics and contextual vulnerability factors.

View Publication

📄 MDPI Sustainability View Paper

Hurricane Persistence and Agricultural Crop Loss

Machine Learning Insights for Agribusiness and Engineering Applications This study examines how hurricane persistence influences agricultural crop loss. The strong predictive importance of lagged crop loss indicates that historical vulnerability plays a central role in shaping agricultural outcomes.

View Publication

📄 ASCE Library View Paper

Enhancing Agile Development

A Comparative Analysis of LLMs for User Story Generation (IEEE DataPort) A comprehensive comparison of various Large Language Models, evaluating their performance, capabilities, and limitations across different tasks and domains.

View Publication

Revealing Landfill Microplastic-PFAS Interactions

Ensemble QSPR Models Predicting PFAS-MP Sorption (Hinkley Center) Research project funded by Hinkley Center for Solid and Hazardous Waste Mgmt. investigating the interactions between microplastics & PFAS in Florida landfills through ML models.

View Project

CuttingEdgeAI

RL for Unique 2D Cutting Stock Problems (Preprint) This project explores the latest advancements in artificial intelligence, focusing on the intersection of machine learning algorithms and reinforcement learning techniques.

View Preprint

Trial

Computer Vision Based Traffic Rule Breach Detection This paper is about the detection of traffic rule breach via computer vision which takes the feed from the traffic surveillance system, processes the video feed, detects the breach and alerts the traffic police.

View Publication

Nazar

Electronic component detection system with server. Electronic component detection, identification and recognition system in realtime.

View Dataset · View GitHub

AI in Project Management

A Systematic Review. This study follows a systematic review methodology. It involves a comprehensive search of literature from multiple databases, screening of studies based on predefined inclusion and exclusion criteria, and synthesis of findings using both narrative and thematic analysis approaches.

View Pre-registration