<feed xmlns="http://www.w3.org/2005/Atom"> <id>https://rgmarelli.github.io/</id><title>Guido Marelli</title><subtitle>Notes on AI systems, agents, and the occasional detour.</subtitle> <updated>2026-08-07T19:00:46-03:00</updated> <author> <name>Guido Marelli</name> <uri>https://rgmarelli.github.io/</uri> </author><link rel="self" type="application/atom+xml" href="https://rgmarelli.github.io/feed.xml"/><link rel="alternate" type="text/html" hreflang="en" href="https://rgmarelli.github.io/"/> <generator uri="https://jekyllrb.com/" version="4.4.1">Jekyll</generator> <rights> © 2026 Guido Marelli </rights> <icon>/assets/img/favicons/favicon.ico</icon> <logo>/assets/img/favicons/favicon-96x96.png</logo> <entry><title>Building a Grounded AI Pipeline: The Seven Stops to a Working Architecture</title><link href="https://rgmarelli.github.io/posts/grounded-ai-pipeline/" rel="alternate" type="text/html" title="Building a Grounded AI Pipeline: The Seven Stops to a Working Architecture" /><published>2026-08-06T00:00:00-03:00</published> <updated>2026-08-06T00:00:00-03:00</updated> <id>https://rgmarelli.github.io/posts/grounded-ai-pipeline/</id> <content type="text/html" src="https://rgmarelli.github.io/posts/grounded-ai-pipeline/" /> <author> <name>Guido Marelli</name> </author> <summary>I wanted one guarantee: every factual sentence the agent produces should be traceable to evidence it actually retrieved, not something the model made up because it sounded plausible. The requirement seemed straightforward, until I tried to make a model with fewer than 3 billion parameters do both jobs at once: answer the question, and correctly cite the evidence behind it. It couldn’t. That fai...</summary> </entry> </feed>
