At Researchly (https://www.researchly.at) we automate market and competitive analysis. We learned that on-the-fly web search through AI agents isn't reliable enough for consistent, in-depth work.
Existing data providers such as PitchBook or Crunchbase are either too expensive or lack the data we need: operational signals such as management changes or new partnerships, for DACH. Furthermore, search tools such as Exa tend to be probabilistic. What they return depends on the model and on SEO rankings, so two runs of the same query can give you two different answers.
So we built our own search layer: several providers (Exa, Linkup, Tavily, Parallel), allow/deny lists, source rankings (we once caught ourselves citing our own blog posts), and checks whether an empty result means "nothing there" or "out of budget".
It works, but: it's slow, expensive, non-deterministic & we kept redoing the same research for different clients (which makes it very un-economic).
So we started building the data layer itself. Step one was Firmendaten (300k companies from the Austrian register; shared here last week https://news.ycombinator.com/item?id=50142669).
Relational data like that is already a huge improvement, but what we really need is a graph, as people ask "W" questions: who does what with whom where and when. This is our first attempt at it.
For now, it just connects who partners with whom and company groups. (Tick "Beziehungen 2. Ebene" to see every partner and person) It's small (a few dozen robotics companies) but you can already see how complex the graph gets & it serves as a good PoC for us.
We had wanted to build this for a long time, and it turned out to be far harder than we expected. One thing that sounds trivial but was particularly hard: finding the right website for a company.
You can buy that data. There are enrichment APIs that return a company's website. But even if they cover Austrian companies, how do you know which company they matched? There is no worldwide unique company identifier.
And if you do it yourself: how do you know that this website really belongs to this company? You have to validate every single match (e. g. through the imprint on a company's website in Austria).
There are quite a few more challenges involved, but we're happy with where this first, admittedly small, result has landed. Feedback very welcome – especially on which relationships would make a graph like this useful for you.
Feedback welcome!
Comments URL: [https://news.ycombinator.com/item?id=50243831](https://news.ycombinator.com/item?id=50243831)
Points: 1