Systems-minded software builder with released, field-tested and beta-ready projects across Windows diagnostics, local AI, computer vision, media automation and evidence-driven data systems. Strong in Python, C#/.NET, SQLite, Git/GitHub, debugging, validation and documenting failure modes. Previous electrical and operations work adds hands-on troubleshooting, quality-control and real-world reliability discipline.
CrashScope · 1.3 in development · 1.2.0 releasedC# / .NET 10 / WPF / WebView2 / React / TypeScript / SQLite · Windows diagnostics
Built and validated a local-first Windows crash-diagnostics system with a native WPF/WebView2 desktop shell, persistent sessions/incidents, Event Log + WER evidence and privacy-safe support bundles.
Reached 266/266 automated .NET tests. The final integrated 1.1 ConfigTrace-off measurement was 0.2365% average Agent CPU, 91.71 MB average / 94.58 MB peak working set and 34.76 MB average private memory. Runtime remains localhost-only with no analytics or automatic upload; CrashScope 1.2.0 is publicly released; 1.3 is the current development line and genuine second-PC validation passed.
Building a native Windows control surface for durable multi-project AI-assisted development via Dev Relay, with guarded execution boundaries, deterministic project activity and persistent Git/result/goal continuity across chats and model changes.
Added a five-provider tool hub, typed product actions, progress/liveness state, recurrence-aware failure learning and source-independent Windows packaging while keeping execution authority and hosted-workflow decisions explicit.
Designed a local-first Windows control deck with GGUF Model Intelligence, safe runtime ownership, hardware telemetry and explainable Smart Launch recommendations.
Added automated update/repair flows, diagnostics, CI, post-update smoke checks and deployment guards that fail visibly when critical state is invalid.
Retro Game VisionComputer vision + decision pipeline · Python / SQLite · Field tested
Built typed data models, provenance-first persistence, compatibility gates, deterministic economics and idempotent validation around messy store-photo/OCR inputs.
Field-tested at 750+ canonical items, 1,000+ physical observations and 1,800+ market observations while preserving uncertainty and evidence lineage.
Built a non-destructive Windows media-library application for legally acquired movies, TV and anime with read-only multi-drive discovery, watch-state continuity, artwork/metadata workflows and source-safe local browsing.
EasyFlix 1.4 adds a shared trusted Add Media path across the GUI, handoff launch and import CLI; the public 1.4 release passed 219 / 219 automated tests with zero build warnings and zero errors, completed public-source privacy sanitization, and has validated installer + portable win-x64 artifacts. The MIT-licensed public source and verified v1.4.0 installer/portable release are live on GitHub.
MARCELO