About
I'm Ariel Alejandro, an AI engineer. I build apps, agents and LLM tooling, mostly in TypeScript, Python and Go, and I publish most of it as open source on GitHub.
Lately I work on two things. One is retrieval for agents: ragtrail, a cache for the search path an agent takes over documents, and how far RAG has come since plain chunking. The other is Fluid, a way of working where coding agents propose and build the work and I decide what ships.
This blog is where I write up what I built, how I measured it and what didn't work.
What I build
Agents and LLM tooling
An embeddable agentic coding library for TypeScript: Anthropic and OpenAI-compatible providers, MCP, built-in tools, sessions and hooks.
A local OpenAI-compatible LLM gateway in Go with real failover between providers and accounts. One small binary, no Docker, secrets in the OS keyring.
A small offline desktop app that turns agent logs and OpenTelemetry spans into a timeline, a replayed chat and a tool-call graph, so you can see what an agent actually did.
How AI agents work from scratch, explained level by level with pixel-art animations and code.
Apps
Keyboard-first access to several LLMs from anywhere in the operating system, using free providers.
Local, private, real-time dictation into any application, powered by faster-whisper.
An animated avatar component that reacts to an LLM chat or voice conversation.
Resources
A directory of LLMs you can use for free or with a generous trial, kept up to date automatically.
A community gallery of ready-to-use prompts for LLMs and generative AI.
Practical study notes for the Microsoft Azure AI Engineer Associate certification.