A working archive
Essays, engineering retrospectives, and notes from the work in progress.
In 2024, everyone pitched a 'digital clone' that would do your chores. By 2026, the question isn't whether AI can act for you, but how much context, permission, and autonomy you actually want to hand over.
AI Town made me realize an AI character did not have to exist only inside an empty chat window. That idea eventually led to Omni Companion: one companion, one living room, and a different way to think about presence.
Most LinkedIn advice stops at attention. This is a look at what comes next: turning a useful post into permission, trust, and a real conversation.
I built an Omni experiment where sending an Instagram DM bootstrapped a guest account and started conversational onboarding. Removing signup friction was easy. Losing control of conversation boundaries was the interesting part.
The story of my first startup, Shopmatey, the lessons learned, and how it led to what I'm building now with Omni.
I used to treat accelerator applications as an obvious habit for an ambitious builder. An interview with Antler Singapore in 2025 forced me to ask a different question: was I building the kind of company that actually needed venture capital?
Publishing an announcement that your startup is joining another platform is easy. The real engineering work is ensuring legacy store links, user identities, and creator content don't break when the old servers wind down.
Letting go of an early brand is hard when you have spent years building it. On April 2, 2025, I announced that Shopmatey was joining Omni. Here is why absorbing what worked into a new product was better than keeping two alive.
Founders often incorporate multiple corporate entities for pragmatic, unglamorous reasons like banking access, fintech tools, and cloud credits. Here is why I had separate entities before consolidating them.
I used to think getting a Google Knowledge Panel was a status symbol for an ambitious founder. When one finally appeared, I realized the panel was just a side effect of doing real work and leaving clean evidence.
I liked building Shopmatey with FlutterFlow. I moved away when owning the backend and keeping infrastructure costs predictable became more important than minimizing the first version.
I applied to an accelerator expecting the exciting part to be getting accepted. What actually helped was having ten structured weeks where learning cloud infrastructure was the job instead of something I postponed.
Most apps make you choose between signing in and signing up. I eventually changed my backend and authentication architecture so the front door became a single button, letting the server figure out whether you exist.
Three days after launching on Product Hunt, I filed a trademark with IPOS. Nine months later, the certificate arrived. Here is what the process looked like, and why owning a trademark is completely different from building a brand.
A retrospective on why FlutterFlow made sense when I was trying to get Shopmatey onto both platforms and still figuring out what the product should become.