Ryan de Melo
I'm a CTO and hands-on builder. Over eighteen years I've taken platforms from zero to production scale: a B2B API business from $0 to $1.1B, ML systems serving 550M+ recommendations a day, and the payments and credit infrastructure behind $22B+ in commerce. These days I build GenAI infrastructure for regulated enterprises.
This is where I write down the parts that don't fit in a slide: architecture decisions, what actually changes at scale, and the unglamorous work of building teams. Start below, or read more about me .
Featured
-
The Support Set Is the Model
Tabular foundation models do their learning inside a context window, which means the ten thousand rows you hand them matter more than which model you picked. Support-set selection is a retrieval problem, and almost nobody is measuring it.
-
Skynet Has a Ticker Symbol
The AI labs talk like they are going to run the world. They do not own themselves. One firm sits near the top of the shareholder register at all of them, votes those shares, supplies the risk model the rest of finance looks through, and is buying the power stations and buildings the whole boom physically stands on.
-
OpenClaw and the Agents That Act: What Non-Technical People Should Actually Know
A plain-English guide to OpenClaw and the new wave of AI that does things instead of just talking. What it is, how it compares to the big-company versions, and the one idea everyone needs before they hand it the keys.
-
I Asked My AI Agent for an Itemized Bill. It Got Awkward.
A month of heavy AI agent use, itemized: only a fifth of the tokens wrote code, almost half could run on a cheaper model, and the prompt cache quietly bills you for every coffee break.
Recent Posts
-
Motto vs Mechanism
American Express held reserves and paid a subsidiary's debts. Amazon lets bad reviews kill sales. Grab and Shopee have mottos too, and the margin squeeze is deciding what they mean.
-
What Eighteen Years of Platform Builds Taught Me About AI Hype
After eighteen years of watching big data, mobile, microservices, cloud, and now AI agents arrive on the same script, here is how I separate the durable capability from the narrative.
-
Building GenAI for Regulated Industries Without Getting Fired
Two ways to fail when you ship GenAI into a regulated business: never ship at all, or ship something nobody can audit. The narrow path between them, from building this in financial services and industrial operations.
-
MCP a Year In: What Held Up, What Didn't
Sixteen months of building production systems on the Model Context Protocol. The interoperability bet paid off. Auth, versioning, and the demo-to-production gap are still where teams bleed.