Predictive modelling & risk scoring
Survival and competing-risks models, gradient boosting, penalised regression. Risk scores, time-to-event forecasts and rankings built to be deployed, not just reported.
Enliva Tecnologías · Málaga, Spain
We build predictive models, causal analyses and uncertainty-aware systems for teams whose decisions carry real consequences — in clinics, breeding programmes and markets.
01 — Services
Most analytics problems are decision problems in disguise. We work on the ones where a wrong answer is expensive — and where the method has to survive scrutiny from clinicians, reviewers or an investment committee.
Survival and competing-risks models, gradient boosting, penalised regression. Risk scores, time-to-event forecasts and rankings built to be deployed, not just reported.
Estimating what a policy, treatment or campaign actually changed. Target-trial style design, propensity and weighting methods, sensitivity analysis — and bounds where the data cannot support a point estimate.
Conformal prediction sets, calibration diagnostics, drift monitoring and validation protocols that hold up: temporal splits, external validation, pre-registered analysis plans.
Registry and cohort data, survival endpoints, genomics and immune-genetics. Auditable pipelines, synthetic data for rare events, and reporting that is ready for peer review.
Genomic prediction, selection indices, trial design and resource optimisation for plant and animal breeding programmes — the engineering behind our BreedIQ platform.
Portfolio construction, risk decomposition, mixed-model and time-series approaches. Backtests with leakage controls and honest transaction-cost assumptions.
Also: NLP and document intelligence, R and Python package development, and reproducible research infrastructure.
02 — Approach
Engagements are scoped in two-week blocks. You get working code and a written record of every choice, so the analysis can be re-run and defended long after we are gone.
What changes if the model says A rather than B? We start from the decision, not the dataset, and agree on what evidence would actually settle it.
Data audit, leakage checks and a validation plan written before modelling begins. If the data cannot answer the question, you hear that early.
A simple baseline first; complexity only where it measurably pays. Uncertainty is reported with every number, and every metric has a confidence interval.
Documented code, an API or dashboard, drift checks and a handover session so your team can own it — including the parts that are fragile.
03 — Work
A small portfolio by design. Each of these is a long-running line of work rather than a one-off deliverable.
A precision-breeding platform for plant and animal programmes: genomic prediction, mate selection and trial optimisation, with the statistical assumptions made explicit to breeders rather than buried in a black box.
Document and record intelligence for organisations with heterogeneous archives. Language models do the first pass; every extracted fact is traceable to its source and review stays with a human.
Portfolio construction and risk analytics used in published quantitative-finance research: mixed linear models for weights, robust risk decomposition, scenario analysis.
Maintained R and Python packages for causal comparison of experiments, optimal experimental design and tabular survival modelling — the same tooling we use on client work.
04 — About
Enliva Tecnologías is a Spanish limited company registered in Málaga. We keep the team deliberately small: the person who designs your analysis is the person who writes the code and answers your questions about it.
The practice grew out of fifteen years spent between academic statistics, a university consulting unit and large-scale clinical prediction work. That background shapes how we work — treat the method as a claim that must be justified, not a recipe that produces a number.
Dr. Deniz Akdemir
Founder & Principal
PhD in Statistics (Bowling Green State University), postdoctoral research in statistical genomics at Cornell University and three years in the Cornell Statistical Consulting Unit. He currently leads clinical prediction modelling in haematopoietic cell transplantation and publishes on causal inference, experimental comparison and uncertainty quantification.
05 — Contact
A short description of the problem and the data you have is enough to start. If we are not the right fit, we will say so and point you somewhere better.