My contribution begins with deciding what software should do and what evidence would show that it does it.
Scope before implementation
I turn a practical problem into requirements, identify constraints, and define the behavior to test. For these projects, that includes source preservation, recoverable jobs, responsive interfaces, and honest failure states.
Direct and review the build
I guide architecture decisions, break work into bounded tasks, and coordinate coding agents. I review changes and behavior, investigate failures, and refine requirements when the evidence reveals a gap.
I do not claim to have personally written every line. AI agents contributed substantial implementation. Libraries, models, codecs, and other upstream work are credited separately in the case studies.
Validate the result
Tests and manual checks serve different purposes. A passing unit suite does not prove a physical player works, and a screenshot does not prove recovery. I keep those distinctions visible, preserve failure evidence, and qualify measurements by their inputs and environment.
Technologies used in these projects
Python, TypeScript, React, FastAPI, SQLite, PySide6, Qt Quick/QML, native mpv integration, FFmpeg, Linux, Git, and NVIDIA media tooling.
This list describes the project environments. It is not a proficiency score or a claim of professional mastery of every technology.
Development tools
My workflows use tools including Codex, Claude, Cursor, Google Antigravity, and Hermes. Local AI work includes configuring Ollama and LM Studio and coordinating GPU resources.
What I make public
This portfolio shares architecture explanations, sanitized demonstrations, and bounded evidence. Application source remains private. The emphasis is on decisions and outcomes that can be discussed and examined.