Aniva Health builds personalised health recommendations from blood test results. I first worked on bringing it to Finland: partnerships, operations and finding the first partners here.
Later I moved closer to the product. I worked on the logic that turns biomarker results into recommendations, and on checking that AI-generated recommendations were accurate and consistent.
The work itself is confidential, so I can’t show it here. What I can say is what I took from it:
- The same lab value can mean different things depending on other markers, medication and the lab’s own reference range. A recommendation has to take that into account.
- Getting a model to write a recommendation is easy. Making sure it’s right every time takes clear rules, test cases and someone reading the results.
- References, safety limits and careful wording take a lot of the time, and they matter.