I’ve been quiet for a while, as the old guard will have noticed. Time for a reveal.
A group of us have been very busy building — with the help of many more amazingly talented folks — and after months of refining the platform with a half dozen intrepid early clients, today we’re officially launching Ragu.AI. It’s a simple, easy-to-use AI platform built on RAG, and I’m incredibly proud of it.
In brief: we help companies and leaders develop their own custom AI assistants from their own company data. That means a more secure, modular and accurate experience compared to ChatGPT.
A word on the acronym, because it’s the whole idea. RAG is retrieval-augmented generation. The term comes from a 2020 paper whose authors observed that large language models carry a great deal of factual knowledge in their parameters, but that “their ability to access and precisely manipulate knowledge is still limited.” Their remedy was to let the model look things up before it answers. Rather than trusting whatever it absorbed in training, you retrieve the relevant documents — yours — and hand them over along with the question.
Samuel Johnson put the principle better than any paper. Knowledge is of two kinds, he said — we know a subject ourselves, or we know where we can find information upon it. A general model has plenty of the first kind about the world and neither kind about your business. RAG supplies the second.
That’s where “accurate” comes from. An assistant that answers from your documents is working from the source, not from whatever it absorbed in training.
“Secure” comes from where the data lives and who gets to see it. OpenAI conceded the point when it launched ChatGPT Enterprise last August, promising not to train its models on business data sent to the service — while noting that ChatGPT had already been “adopted by teams in more than 80% of Fortune 500 companies.” Put those two facts side by side. The demand was inside the building long before the guardrails were.
“Modular” is the plainer virtue — pieces that come apart, so that changing one needn’t mean rebuilding the rest.
