Rag

Roz answers questions about me in the visitor’s language, from what this website publishes and a set of notes written for her, with links to where each answer comes from. This is how she is designed: the idea behind her, her architecture, the search that decides what she reads, and the rules that keep her honest.

A retrieval system over four chapters of the PostgreSQL documentation. Both of the bugs that mattered were silent: a parser that returned the right number of chunks and left every one of them empty, and an out-of-scope threshold that confidently answered questions whose answers lived in other chapters. Neither raised an error, and each needed a different kind of test to surface.

A conversational agent for e-commerce built with Ollama, LangChain, and ChromaDB, comparing three open-source LLMs — Qwen3 4B, Llama 3.2 3B, and Phi-4-mini — for product search, order tracking, recommendations, and RAG-powered FAQ. Benchmarks tool-calling accuracy, multi-intent handling, and response time across all three models.

‘Gabo’ is a RAG (Retrieval-Augmented Generation) system designed to enhance the capabilities of LLMs (Large Language Models) such as ‘DeepSeek-R1’, ‘Llama 3.2’, and ‘Phi 3.5’. This project honors Colombian author Gabriel García Márquez by marking the tenth anniversary of his death.