Selected Work

Multi-agent AI co-pilot for 911 dispatchers, built on Amazon Bedrock. Six specialized agents (Claude Haiku 3.5 + Claude Sonnet 4) process real-time Amazon Transcribe streaming, run parallel domain watchers in AWS Lambda, and surface a single dispatcher recommendation in under 15 seconds — grounded in FEMA and MPDS protocols via Bedrock Knowledge Base with RAG.

Peekwise.ai

Monorepo competitive intelligence SaaS — BullMQ as the sole async handoff between ingestion, AI enrichment, and orchestration layers; cost-zero Groq pipeline processing ~2,075 reviews/day; Redis-first sub-millisecond dashboard reads.

Multi-agent code review platform in C#/.NET 8 on Microsoft Semantic Kernel — four specialist agents run concurrently for security, performance, and style analysis, fronted by a versioned ASP.NET Core REST API and a Recharts quality dashboard.

Next.js 16, React 19, TypeScript strict, Tailwind CSS v4 — MDX blog pipeline, 8-section neo-brutalist design system, PostHog analytics, and Claude Code AI-assisted development workflow deployed via Vercel.

Habiwine

Production SaaS backend with Python & Node.js — RESTful APIs, PostgreSQL optimization (30% faster queries), Cloudflare edge caching cutting repeat load by 35%, and AI-powered habit analytics.

Signal Protocol–based end-to-end encrypted chat demonstrating zero-knowledge messaging — X3DH key agreement and Double Ratchet implemented from the official specs in TypeScript with TweetNaCl, real-time WebSocket delivery, and an encrypted store-and-forward queue for offline recipients.

Client–server image processing service in Python using gRPC and Protocol Buffers — single-RPC API contract for binary image transfer, seven chained operations in one round-trip, concurrent 10-worker server with Pillow pipeline, and a defensive interactive CLI client.

Collaborative mental wellness journaling app built with Angular and Node.js/Express REST API backed by MongoDB — full CRUD for mood-tagged journal entries, Dashboard View All, and Mocha/Chai test coverage.

Comprehensive automated test suite for an open-source Java Swing game, built for a Software Testing course at Seattle University — 104 tests across 17 classes covering a real-time application with moving objects, timers, and rich UI interactions, hardened with mutation testing and branch-coverage analysis.

ML fake news detection achieving 96.8% accuracy with Random Forest Classifier — NLP with TF-IDF vectorization, Word2Vec embeddings, and deep learning models (Neural Networks) for classification.

Ethereum smart contracts in Solidity securing IoT device communication — 40% reduction in unauthorized interactions, Python APIs for blockchain interactions, and MongoDB-backed distributed data workflows.

Predictive ML models with Python & scikit-learn achieving 25% better risk assessment accuracy — Kafka streaming for large-scale data ingestion, MySQL + Memcached query optimization, and 20% reduction in false positives.