Everybody is familiar with chatbots today. While they can be powerful, their answers are not always reliable. So how can AI be used to search complex, highly structured data sources like Biotechgate without losing precision?
Do you know how most AI-powered search tools actually work? At their core, many of them rely on semantic search. Your query is converted into a vector, and the system returns results that are similar in meaning, but not necessarily exact matches. This works well for documents and general knowledge. However, when it comes to structured data, similarity is often not enough, because similarity does not equal precision.
When we decided to start using AI for search in Biotechgate, we did not want to just build another chatbot on top of our data. Instead, we focused on a different goal: Making complex, structured datasets accessible through natural language — without losing accuracy. The result is a different kind of AI search because our system translates natural language into structured queries instead of returning “similar” results.
Take a query like:
“Swiss biotech companies with financing rounds in the last five years that have licensed assets to the US”
Behind the scenes, this is not treated as text, but rather converted into a precise combination of filters:
- country = Switzerland
- sector = biotechnology
- financing activity with a time constraint
- licensing activity in the US
These filters are then combined into a deterministic query that returns exact matches rather than approximations. Additionally, you can see which search filters were applied.
This becomes even more important with more complex queries, like the one below:
“Companies with a lead asset in Phase II, a market cap above USD 50 million, excluding oncology”
This type of query combines multiple dimensions: Development stage, financial thresholds and explicit exclusions. Traditional semantic-powered searches struggle here because they can find “similar” companies, but cannot reliably apply strict constraints — especially when it comes to exclusions.
Our approach can handle this as it operates on structured filters, not similarity. This is where structured data provides a real advantage. Our data is not just text, it is organized into defined categories and taxonomies — from therapeutic areas to indications — as well as standardized attributes such as country, sector, and financial metrics.
This structure allows us to combine filters in a precise and reproducible way while also enabling something that many AI systems lack: transparency. Instead of acting as a black box, our AI search shows how a query was interpreted: which filters were applied, how they were combined and why specific results were returned.
Transparency makes it easier to trust the results — and to refine queries when needed. In the end, the difference is simple: most AI search tools help you find similar things, while our natural language search leads you to the correct results. At Biotechgate, we know that AI is not just about understanding language. Instead, its real value lies in connecting language with structured knowledge, which makes the difference.
We are currently testing this new search experience and continuously refining it based on real user queries. The feature will be launched on Biotechgate in Q2 2026.