Building the
agentic layer
of software.
I'm Thiago, an AI Product Engineer designing autonomous agents, RAG pipelines and spec-driven products that ship fast and learn faster.
- 12+
- Years shipping
- 106
- Articles
- AI · TS · RN
- Stack
Field notes from the edge.
Deploy cancelled by Out Of Memory: WiredTiger cache in containers
A Dokploy deploy stopped during the Next.js build and was cancelled without a clear error. The cause was memory pressure: two MongoDB instances in containers calculated their WiredTiger cache from the VPS RAM, not from an isolated container limit.
RPC, REST and GraphQL: the same remote call with different interfaces
In a fullstack TypeScript web app, REST, GraphQL and RPC/tRPC look like three different worlds. But all three serialize data, send it over HTTP and read the response back. They all live on OSI layer 7. What changes is not how remote the call is, it is the mental interface at the top, and the price each abstraction charges. With the tRPC vs Server Action contrast on the same create.
The harness is the product
WorkOS, Stripe, and OpenAI arrived, through different paths, at the same conclusion about coding agents: senior engineering stopped being about writing code and became about building the harness that makes the agent write code reliably.
Where Java with Spring Boot still beats Node.js (and where it doesn't)
It is not Java vs Node in the abstract. It is about where the maturity of the Spring ecosystem solves the boring production problems out of the box, the ones Node still puts together piece by piece, and where Node + TypeScript stay the obvious call. With honesty about Virtual Threads, types that vanish at runtime, and what really decides: the requirement.
Back to the fundamentals: I'm taking Andrew Ng's Machine Learning Specialization
Why, after spending every day building with LLMs and agents, I decided to go back to classic ML basics — and what each of the three courses in Andrew Ng's specialization actually covers.