Domain Resolution Needs Reasoning Not Rules

Your CRM thinks it has clean data. It doesn't.

And no string-matching algorithm is going to fix that, because the problem was never about strings.

• getstripe.com and stripe.com are the same company, just different marketing domains.

• aws.amazon.com is a subsidiary of Amazon, but careers.google.com is not, even though the same logic would say it is.

• novonordisk-us.com is Novo Nordisk operating in the US, not a separate entity.

• A bit.ly link gives you no company information at all until you follow it.

And the CRM data underneath all of this is already messy, full of wrong domains, duplicates, and entries that passed whatever basic check was in place at the time.

The real question is not a string match, rather, do these two domains belong to the same company?

That question needs judgment, not logic. It needs reasoning!

So the system we built reflects that. It resolves domains to DNS, scrapes homepages, uses an LLM to compare content and make a call on company identity, adds web search to reason about subsidiaries, and uses heuristics and world knowledge to generate the right candidates before any of that runs.

The architecture is layered because the problem is layered.