NVIDIA teaches AI what your business words mean

NVIDIA's open-source Auto Ontology gives AI agents a shared dictionary of your business, so they guess less when they answer questions from your data.

A small problem with big consequences

Ask three people in your company what "active customer" means and you'll probably get three answers. Sales counts anyone who bought this year. Finance counts anyone with an open invoice. Now picture an AI agent answering "how many active customers do we have?" by reading your databases. It quietly picks one meaning and hands you a confident number.

What NVIDIA built

Auto Ontology is a free, open-source project from NVIDIA that sits between your data and the AI. It reads how your systems describe things, lines up the definitions that clash, and builds a map of what words actually mean in your business.

What I like is the order of things. The tool drafts the map, real people check and approve it, and only then does the agent use it to answer questions. Humans decide the meaning. The machine does the legwork.

It works with any AI agent, through MCP or other routes. A companion project measures how much more accurate the answers get, and lists which questions it fixed and which ones it broke. That kind of honesty is rare.

The catch

It's heavy. One independent review counted 129,285 lines of code, and it needs your own NVIDIA API key plus a specific database called Postgres. This is a project for your tech partner, not a weekend toy.

Words worth knowing

Ontology: a shared dictionary of your business terms and how they connect.
Text-to-SQL: AI turning your plain question into the language databases understand.
MCP: a common plug that lets AI agents connect to tools and data.

Something to try this week

Before any AI touches your numbers, write down your five most-used business words and ask each team what they mean. That conversation might be the most valuable part.