I’m Rafael, a software engineer. Here you’ll find some of my writing and projects. I hope you enjoy them.
A local retrieval-augmented generation lab covering document ingestion, vector search, grounded answers, and reproducible evaluation.
A NestJS API for CSV imports that checks uploaded files, stores them privately, and records each import in PostgreSQL, with a modular path to more formats.
An event-driven automation backend with signed webhooks, asynchronous workers, durable execution history, retries, and dead-letter replay.
A reading platform covering libraries, collections, reading sessions, and community spaces, presented through an interactive UI showcase.
What a fixed Sonora benchmark reveals about tool use, grounded answers, model pass rates, judge bias, and estimated costs.
A practical guide to loss functions, gradients, learning rates, mini-batches, backpropagation, modern optimizers, and Python examples.
A practical comparison of fixed and agent-driven retrieval, with original diagrams, an execution simulator, and the engineering trade-offs behind both designs.
A story about how a model learns from noisy heat-flow measurements, what fitting a line accomplishes, and why useful predictions need more than a good fit.
A measured Qwen3 4B and 0.6B comparison on a 6 GB RTX 3060 laptop GPU, covering generation speed, VRAM use, and answer quality in a small RAG test.
I’m Rafael, a software engineer with a degree in electrical engineering. I’ve worked on production-scale systems used by real users, building web applications, APIs, payment integrations, and asynchronous workflows with Python and TypeScript.
My focus is backend engineering, cloud infrastructure, and applied AI, with independent projects and open-source contributions. I’m deeply interested in machine learning and large language models (LLMs). Outside of software, I enjoy music. Here I share what I build and learn.