How Iris works

What a financial knowledge graph is, and why Iris is built on one

Akshith Hiremath · October 18, 2026 · 5 min read

Most tools that watch your portfolio work by matching names. If an article mentions your ticker, you see it. If it does not, you do not. That misses a whole class of news: the story about your company’s supplier, its biggest customer, or the competitor it moves with.

What a knowledge graph is

A knowledge graph is a map of things and the relationships between them, stored as data. In Hivemind’s graph the things are companies, their subsidiaries and the private and foreign suppliers behind them: more than 14,000 of them. The relationships are typed and directional: this company supplies that one, this one is a customer of that one, these two compete. There are more than 42,000 of them, and each keeps the evidence it came from, such as the filing or disclosure that established it.

Because the relationships are stored rather than guessed each time, they can be walked. Start from any company in a news story and follow its edges outward, and you reach the companies one or two steps away.

A real example

A path in Hivemind’s graph, October 2026

Ajinomoto→Ibiden, Unimicron, Shinko, AT&S→Intel, NVIDIA

Ajinomoto is best known for seasonings. It also makes the build-up film that advanced chip substrates are made from, and the substrate makers that use it supply the largest chip companies. A story about an Ajinomoto plant will never mention a chip stock, so ticker matching will never connect it to your NVIDIA position. In the graph it is two hops away.

A path like this shows how a story can reach a holding. It is not a forecast that the stock will move, or by how much.

Why not let a language model work it out?

Large language models are very good at reading and explaining. They are less reliable at finding an obscure relationship nobody asked about, and they can describe one that does not exist with total confidence. Ask a model how Ajinomoto relates to NVIDIA and it may well explain it. It will not reliably think to ask on its own, every morning, for every holding.

So Iris splits the work. Models help read filings and news into structured facts, and write the explanation you read. The graph decides what connects to what. We think language models are most useful working over a structured, traceable map of the market, not as the map itself.

What it means for you

  • Stories that reach your holdings through a supplier, customer or competitor show up next to the ones that name them.
  • Each one comes with its path, so you can see why it is there and decide whether it matters.
  • Your funds are opened up too: Iris maps the companies inside an ETF from its N-PORT-P filing.

You can try it on your own tickers, without an account, on the scan page.

Akshith Hiremath, Co-founder & CTO

Built the knowledge graph architecture and the news ingestion and processing pipelines that route each story to the holdings it reaches.

See what reaches your portfolio.

Run your own tickers through the map in ten seconds, no account. Or start the 14-day free trial, no card required.