I build systems that turn intelligence into action: autonomous agents, research pipelines, remote coding infrastructure, and small products that make messy workflows usable.

• Agent operating systems: memory, tools, queues, verification, and feedback loops for autonomous work.
• Local-first AI products: software that keeps context close, useful, and owned by the user.
• Remote coding infrastructure: workstation workflows that survive laptops, phones, bad Wi-Fi, and long-running agent jobs.
• Productized research: turning market/technical sensing into shippable bets instead of dusty notes.

• openclaw: Linux saved Gateways stay paired across restarts instead of asking for approval every launch.
• ruff: SLF001 stops flagging private access on object.__new__(cls) instances.
• webdriverio: multiremote mockRestoreAll() restores only its own browser's mocks.
• x402: bazaar discovery keeps the declared HTTP method instead of copying the probe's.
• mathlive: focus no longer breaks after a focused mathfield is removed from the page.

• Ori Notebook: local-first notebook where agents work inside your notes.
• Sigira: AI film studio. Every piece shows the model calls and dollars behind it.
• Night Circuit: drive Buenos Aires at night in the browser, every street and building from real city data.
• Lobby: web game platform, five cross-play games live.
• Proof: analytics for agent-built apps.
• dorsr.com: proof-of-work site and the Model Atlas.

More serious repos are being cleaned up and promoted from private work into public artifacts.

• agents need memory, verification, and taste, not just prompts
• private work should compile into public proof
• small tools beat vague roadmaps
• if it does not ship, it is just expensive journaling