LLM's aren't often trained on hacking and redteam data, but they're great at systems knowledge and writing code. A RAG helps augment an LLM with new information that it wasn't previously trained on. It can then blend the new information with its existing knowledge to produce interesting results.
VX Underground has a huge archive of papers and malware code examples, but no search engine.
vxllm uses ColBERTv2 for embedding, retrieval and reranking. The Ragatouille project makes implementation much simpler than the original project.
BERTScore evaluation: