MICHAEL CZEISZPERGER

Software Architect | Agentic AI Architecture | AI-Assisted Development

I've been building developer tools for 25 years, most recently modernizing a 559,000-line Java codebase and integrating AI agents directly into the product: 67 MCP tools, agentic test configuration, and automated report generation. The modernization itself was done using Claude Code and multi-LLM spec-driven development. Clients have included the US Bureau of Economic Analysis, the New York Marathon, the US Census, the Governments of Canada and France, and scores of others. I got my start doing R&D for Yamaha Music, then worked on Solaris at Sun Microsystems, and have been building and shipping my own software ever since.

Professional Work

At Web Performance, I've served enterprise clients including the US Census, Canadian government, and New York Road Runners.

Legacy Modernization

Modernizing a 559K-Line Codebase

Modernized a 25-year-old, 559,000-line Java codebase as sole developer using Claude Code and multi-LLM spec-driven development. Java 6 to 11, Eclipse RCP 3.6 to 4.19, SWT to React 18, Ant to Maven.

  • Java 6→11, Eclipse RCP 3.6→4.19, Ant→Maven, SWT→React 18
  • 559K lines modernized as sole developer
  • 1,294 commits with 791 automated tests
  • React 18 dashboard with JWT auth and containerized CI/CD
  • Multi-model AI consensus for spec-driven migration
Agentic AI Systems

Agentic AI for Load Testing

Built agentic AI systems for Web Performance Load Tester 7.0: a test configuration agent with 75 MCP tools and a 3-stage router, agentic report generation with an AI-driven investigation loop, and structured RAG using triage hierarchy, recipes, and live data retrieval.

  • 75 MCP tools for autonomous test configuration and analysis
  • 3-stage router: UI context, intent classification, tool selection
  • Agentic report generation with AI-driven investigation loop
  • Structured RAG: diagnostic triage, recipe system, live data retrieval
AI-Native Development

WPLoadCompare

Built from scratch with Claude Code. Customers needed side-by-side comparison of load test results across different test runs, so I built it, spec to signed installer.

  • Spec and test-driven development with Claude Code
  • Java 21 Eclipse RCP/SWT
  • Integrated with existing Web Performance ecosystem
  • Signed installers for macOS and Windows.

Personal Projects

Mobile App

WalkOnAlerts

iPhone/Android app that sends real-time notifications when theme park rides reopen after breakdowns, helping visitors catch walk-on opportunities before crowds return. Built on breakdown data from 86,500+ analyzed events across 31 US theme parks.

  • React Native / Expo with TypeScript; Python/Flask REST API backend
  • Scikit-survival gradient-boosted models with 23-feature vectors for breakdown duration prediction
  • Tiered push notifications via Expo Push API with deduplication and batching
  • Sub-60-second reopening detection across 31 US parks
  • Data-driven stay-or-leave recommendations using survival curves at 97 time-point intervals
Agentic AI/MCP

AI Financial Analysis System

Natural-language financial reporting for Copilot Money. Reverse-engineered the GraphQL API and built a Chrome extension with 12 specialized tools that answer complex analytical questions about your spending.

  • TypeScript/React 18/Vite Chrome Extension (Manifest v3) with MCP tool pattern in service worker
  • 12 specialized tools with branded types for compile-time USD safety
  • Deterministic math via dedicated calculator tool, AI never touches arithmetic
  • Levenshtein-distance fuzzy matching with 4-strategy confidence scoring for category resolution
  • Multi-model AI (Claude + GPT fallback) with autonomous error recovery
Open Source

ScrollKit

A niche labor of love: a CircuitPython library for driving scrolling LED displays showing real-time data from anywhere. Powers the physical LED product sold at ThemeParkWaits.com.

  • Three-process async architecture: display (~50 FPS), data updates, and embedded HTTP server
  • Pygame-based LED matrix simulator with factory pattern for platform abstraction
  • OTA live code updates via GitHub integration with circuit-breaker restart protection
  • Embedded web UI (Adafruit HTTP Server) for WiFi config, park selection, and brightness control
  • Pre-allocated fixed-size queues and explicit GC to manage ~200KB ESP32 RAM