Skip to content

Brief

A retrieval-augmented assistant running a fine-tuned Llama entirely on local hardware, with an agentic workflow over several tools. Evaluated on more than 1,000 prompts and measured at 92 percent accuracy on drug-related queries.

All projects