BlueberryX AI
BlueberryX builds AI pricing engines for insurers. Gradient boosting replaces stale GLM tariffs, with per-quote explainability and EU compliance documentation built in, delivering 3 - 5 points of loss-ratio improvement, proven on the insurer's own historical data. Romania first, then CEE.
BlueberryX AI
BlueberryX AI
Romania, Bucharest
Founded by: Vlad Rosca - Co-founder. Product, engineering, strategy.
Second-time founder. Previously built and sold a US startup in drone-based inspection for landfills and solar farm which was bootstrapped with no outside capital and acquired within 18 months of launch. Studied at UCLA and UC Berkeley. At BlueberryX he owns the engineering and product side: the pricing engine, the explainability layer, and the data architecture the whole thing has to be defensible on.
Andrei Stefan Radu - Co-founder. Insurance domain, commercial.
Has spent his entire career inside insurance companies, including time at the Bank of England. Educated in the United Kingdom and Switzerland. At BlueberryX he owns the domain: what actually drives loss behaviour in Romanian motor business, how carriers buy, and the relationships with them. The technical decisions in the pricing engine get argued against his experience before they ship.
Insurers in Romania and Central Europe still price motor business with GLM tariffs built years ago and refreshed rarely. The mispricing is systematic, measurable and expensive. BlueberryX does one thing about it: pricing. We build machine-learning pricing engines - gradient boosting trained on the insurer's own claims history - wrapped in the layer that makes them usable under EU supervision: exact per-quote explainability, EIOPA differential-pricing documentation, and AI Act technical files as product features rather than afterthoughts. We deliberately exclude demand and renewal-price elasticity. Our models price risk, not a customer's willingness to pay. Founded by two friends: one from inside the insurance industry, one a second-time founder. Romania first, then CEE.
Facts and numbers
Key facts
- 1-5 Employees
- MVP development stage
What we have achieved:
BlueberryX has built the canonical quote schema its pricing engine runs on: Romanian motor business - RCA, CASCO and fleet - around 208 fields, each carrying a rating-eligibility class, a product-line tag and a missingness policy. Machine-validated JSON Schema, with a continuous-integration test that mechanically blocks identity fields, outcome variables and non-rating data from reaching the model. Gender, nationality and GDPR Article 9 categories have no fields at all.
Building it produced a finding. We rebuilt the CAEN Rev.2 to Rev.3 crosswalk from all 651 codes and hit a live data trap: code 4932 meant "taxi" before 1 January 2025 and means "occasional passenger transport" after it - taxi moved to 4933. Any insurer backtest spanning 2022 to 2026 silently merges taxis into coaches, in the segment where mispricing costs most.
Engineering runs in two private repositories, continuous integration green. The compliance architecture - EIOPA differential pricing, EU AI Act documentation, exact per-quote explainability - was settled before the first model is trained, not retrofitted after.
Self-funded. The schema freeze is the last gate before the engine build.
Our interests:
Funding, events, startups, accelerators, people, networking