Clear lessons on AI workflows, engineering, and digital products
Large language models are stateless. Every time you send a request, the model does not truly “remember” what happened before. It simply receives a new input containing the system prompt, part of the conversation history, and the latest user message. And yet, tools such as Claude Code can appear to
Enterprise RAG sounds simple on paper: connect documents, retrieve relevant context, send it to an LLM, and return an answer. In practice, that is not what separates a convincing demo from a system a business can trust. A real enterprise RAG solution has to deal with changing data, fragmented tools,
The AI agent framework market has gone from a Cambrian explosion to something much harsher: a survival test. In the first wave, almost every new framework looked exciting because almost every demo looked magical. By 2026, that phase is over. Teams are no longer choosing based on who can produce
So far in this series, we have built a NestJS API that can: * define routes * validate input * organize logic cleanly * return JSON responses * feel like a real backend Very nice. But there is still one big problem: the data is probably living in memory Which is totally fine for learning.
At the beginning, a NestJS app feels pretty manageable. You have: * a few modules * a few routes * some DTOs * a service or two * maybe a little auth * maybe some Swagger * and life feels reasonably calm Then the app grows. And suddenly: * modules start depending on each other too much * business
