I'm a backend and systems engineer in Buenos Aires, currently at @trysandbar. Most of what I build sits close to the metal: a message broker, a file index, some local dev tooling. I put the benchmark table in the README, because a speed claim you can't check isn't worth much.

Before Argentina I studied CS at UTEC in Peru, which is still where most of my instincts about data structures come from.

What I'm building

• chasquimq is a Redis-backed message broker and job queue with a Rust engine, MessagePack payloads, and native Node and Python bindings, so handlers run in the language you wrote them in. It reaches 188,775 jobs/s on bulk enqueue, 3.47× BullMQ on the same machine. Published to crates.io, npm and PyPI, with docs at chasquimq.io.

• unport runs Docker Compose projects at real hostnames like https://api.shop.test instead of a pile of localhost ports. It holds two ports for the whole machine, writes one generated override file, and leaves docker compose up working the way it always did.

• prism is filename search for macOS, built the way Everything works on Windows. DuckDB holds the metadata, SQLite FTS5 holds the index, and a memory cache sits on top: about 8 ms to search 27,604 indexed files, and it still answers when the external drive is unplugged.

• pdse answers questions about multi-hour podcasts with clickable timestamp citations. I built the instrument panel before the features, so every search emits a trace carrying its token cost, retrieval scores, and grounding signal.

• qail reads a cold email inside Gmail and scores it before you send: spam risk, tone, personalization, and the reply you are likely to get back.

Around those languages, the tools I keep coming back to: Redis, Postgres, DuckDB and SQLite for storage, Docker for anything local, and enough CI to keep a benchmark honest between commits.

Elsewhere

• chasquimq.io for docs and writing about the broker
• @jotarios_ on X
• j29.rios@gmail.com if you have a problem with a latency budget

_{The numbers above are regenerated weekly by a small workflow that queries the GitHub API and commits the SVG, so nothing here depends on a badge service staying online.}