Notes
Things I'vebeen thinkingabout.
Decisions I made and why, what I learned the hard way, and what I took away from people worth listening to. Mostly about building, robotics and getting things into production.
The adapter was the bottleneck, not the SLAMWe spent weeks optimising a SLAM pipeline whose real limit was the layer translating between two message formats. The lesson is about where you point a profiler, not about SLAM.We took out the auth plugin and wrote our own RBACBetter Auth's organizations plugin models a flat company. A restaurant group is a hierarchy, and permissions have to inherit down it while never escalating up. Bending the plugin would have cost more than replacing it.Why Awake's helper is a LaunchDaemon and not a sudoers ruleThe whole point of the app is asking for a password once instead of every time. Only one of the two options survives a reboot, and that decides it.Seven green deploys that never shippedRestoman's API served the previous night's build for a day while seven deploys reported success. A stuck migration and a deploy command that exited 0 on failure hid each other.We went to learn someone else's tool and left with a class to teach at ITBAA Google hackathon at ORT was about making videos with Google Vids. We used the room to show Gemm Blocks instead, which led to an invitation to teach it at ITBA.Two hours to design triage—and most of the work was defining limitsA hackathon at Di Tella asked how AI could help where the nearest specialist is six hours away. Our answer was defined less by its features than by the lines it would not cross.What I took away from Daniel Rabinovich's talk at Mercado LibreWith ORT, we visited Mercado Libre's offices. Four ideas on the impact of artificial intelligence and the future of talent, and a call to get in the game.