When organisations say they manage knowledge, they usually mean they store documents.
I built VirgiLLM to pick a fight with another windmill. Organisations write documentation, forget it, and move on. Over time, what is supposed to be the soul of your company ends up forgotten, full of contradictions, and festers with the kind of imprecision that is easy to exploit for bad actors.
VirgilLM tries to fix that by providing an enterprise knowledge platform using three specialised analysis pipelines. Together, they provide a golden document that you can feed to any RAG and interrogate.
I uses Rust and NLI and NLP and LLMs and so many acronyms. So many.
The design philosophy is simple: use each technique for what it is actually good at. Vector embeddings excel at finding candidates; LLMs reason about contradictions. The system uses k-NN to narrow results, then LLMs to verify, with a sceptical second-opinion LLM controlling precision.
Virgil is agentic AI built on craftsmanship instead of vibes. It's too early to say how well it will do, but I have a good feeling about this one.
