At SilkForge AI, I work on independent software projects that take an initial problem through implementation, debugging, testing, and documentation. For Grid AI, I built an Electron workspace with a Python backend, streaming chat, file editing, terminal output, and change review.
I added safeguards to Grid AI that prevent skipped or zero-test runs from being marked verified, along with Git checkpoints, task undo, and regression rollback. I also integrated cloud and local model providers and MCP tools with approval controls and visible risk information.
On ADOM, I built controls for agent permissions and execution, including scoped capabilities and recoverable SQLite state. In an author-run concurrency test, it blocked 226/226 unsafe cross-session actions while allowing 480/480 benign actions; on Rain, I kept model-generated claims separate from deterministic checks and retained unsupported claims as unverified.
