Dylan Castillo

As people began using LLMs with their own documents, a new problem emerged: how do you store and search all that information efficiently?

For the past few years, Simon Willison has tested every major LLM release with the same prompt: “Generate an SVG of a pelican riding a bicycle”.

Five years ago, freelancing felt like arbitrage: same work, better pay, without leaving Spain.

I thought I was seeing fewer arXiv papers on the front page of Hacker News (HN) these days, and I wanted to check if that was real.

While I use Claude Code to build crappy forms, real hackers use it to make money.

I’m an independent AI consultant, trying to grow my business.

A simple hack that will make your life easier at work: Always cover your ass (CYA).

The best thing about AI is that EVERYONE can build now.

To make Shift+Enter create a new line in Claude Code when working with Alacritty/Zellij, you need to add the following lines to ~/.config/alacritty.toml: [keyboard] bindings = [ { key = "Return", mods = "Shift", chars = "\n" } ] I found the solution…

Hi, I’m Dylan, and I’m a ClaudeCodeholic.

In my first AI projects, I didn’t have access to proper observability tools and didn’t know how to evaluate the performance of LLM pipelines.

One of the coolest ideas I’ve come across when building Retrieval-Augmented Generation (RAG) systems is that you can actually start without any real data at all.

I’ve been re-implementing typical patterns for building agentic systems with Pydantic AI.

I’m doing a deep dive into Pydantic AI, so I’ve been re-implementing typical patterns for building agentic systems.

I’m trying to get more familiar with Pydantic AI, so I’ve been re-implementing typical patterns for building agentic systems.

To get more familiar with Pydantic AI, I’ve been re-implementing typical patterns for building agentic systems.

I wanted to get more familiar with Pydantic AI, so I decided to build a Reasoning and Acting (ReAct) agent with multiple tools.

As Large Language Models (LLMs) have become more powerful, I’ve started to see increasing interest from clients in building agents.

These days, everyone is “building” agents.

Function calling and structured outputs let you go from chatbots that just talk to agents that interact with the world.