Mapping deep tech through patent embeddings
Turning thousands of patent filings into an interactive landscape — so you can find the whitespace before someone else does.
Patent data is one of the best signals for where a technical field is actually going. It’s also mind-numbing at scale. A serious technology area can have thousands of filings, and reading through them to understand the landscape is a weeks-long project before you’ve even started the real work.
I’ve built a tool that takes a patent search query and turns it into an interactive landscape. It pulls up to 5,000 patents from the European Patent Office, generates embeddings from their abstracts, and clusters them by technical similarity. What you get is a visual map where proximity means technical relatedness, not just shared keywords.
The part I find most useful is the whitespace view. It maps filing density across technology clusters, so you can see where activity is heavy and where it’s sparse. When you’re scoping out a new domain, that’s the view that tells you where the opportunities might be.
You can also investigate technology categories through a timeline view to see which areas are heating up or cooling off — useful to know if you’re walking into a crowded space.
Every view shares the same selection state, so narrowing to a single company or cluster immediately shows you how that slice relates to everything else: where they’re filing relative to the crowded areas, and where they’re not.
Seven pre-computed landscapes ship with the tool so you can explore immediately. The additive manufacturing one makes obvious how concentrated filing activity is around a few process types, and how thin it is in areas like in-process monitoring and multi-material — which are where I find most of the interesting problems are.
If you want to investigate your own areas, you can connect your free EPO API key to run custom searches.