Groundball is a local-first baseball analytics assistant that answers MLB history questions with DuckDB-backed facts, visible SQL, provenance metadata, stat-claim verification, and deterministic release gates.

What it demonstrates:

• Grounded generation: LLM prose is constrained by typed routing, source rows, and dataset manifests.
• Eval-gated releases: deterministic CI checks guard supported questions, unsupported cases, SQL visibility, and provenance coverage.
• Audit-ready answers: responses expose SQL templates, row evidence, checksums, license metadata, route metadata, and review queues.
• Local-first AI workflow: Python, FastAPI, DuckDB, Typer, Gradio, pytest, and local/open-weight model integration.

Repo: https://github.com/DiscoStew6082/groundball

A SwiftUI/AppKit desktop editor that keeps Markdown as plain text, builds a Finder-launchable .app, and treats hostile file intake as a first-class design boundary.

What it demonstrates:

• Native macOS UX: SwiftUI app shell with AppKit NSTextView editing behavior.
• File-safety boundaries: rejects symlinks, special files, oversized files, NUL bytes, and invalid UTF-8 before loading content.
• Tested desktop edge cases: Swift Testing coverage for document state, stale async opens, save flows, and paste behavior.

Repo: https://github.com/DiscoStew6082/ObviousSaveMD

Building practical AI systems that make model behavior inspectable: grounded answers, reliable fallbacks, provenance, evals, accessibility, and local-first automation.