We research beforeanyone asks for it.

Olive oil is one of the most volatile agricultural markets in the world. The question was straightforward: how predictable is it, and with what data?

We spent several months on successive iterations, expanding sources and testing different model families on horizons from 1 to 52 weeks. The most valuable conclusion was not the best-performing model — it was understanding why certain approaches fail and where the real limit of forecasting lies in this kind of market.

At short horizons, price direction is predictable with useful accuracy. At long horizons, the market reacts to external shocks no model can anticipate consistently. We publish these results because honesty about limits is part of serious research.

Standard satellite imagery is not calibrated for Mediterranean crops. Cloud cover, dry-soil reflectivity and the specific phenological cycles produce noisy readings when generic methodologies are applied.

We researched how to build a spectral analysis system that correctly distinguishes vegetative vigour, water stress and soil status under high insolation and low humidity. The work included exploring band combinations that standard indices do not cover.

The result is a processing system that operates at two scales — regional and per parcel — with different architectures depending on the use case. The efficiency gap between both is three orders of magnitude.

Urban digital twins demand precise geometric data for thousands of buildings. The usual problem is that this data either does not exist or sits behind private licences that make the project impossible.

Spain — and Europe in general — has high-quality open data coverage that almost nobody exploits systematically. We researched how to turn those official sources into navigable 3D models without depending on any proprietary data or external provider.

The outcome is an architecture that can be replicated in any municipality with official coverage. The geometric accuracy obtained is comparable to high-cost commercial solutions.

The market intelligence problem is not lack of information — it is excess. Dozens of sources publishing about the same events with different angles, languages and relevance.

We researched how to build a system that automatically clusters related information, scores its relevance for a specific context and generates coherent syntheses without human intervention. The main technical challenge is not generation — it is coherence validation before content reaches the user.

The architecture is domain-agnostic: it works for any vertical with continuous unstructured information flow, from financial markets to regulatory tracking or competitive intelligence.