Explainable AI
Making ML models transparent for clinical and sports professionals: SHAP for feature contributions, LIME for local interpretability, Grad-CAM for visual attribution, counterfactuals for actionable what-if reasoning.
Research & AI · alongside the engineering
Explainable AI for human locomotion
A PhD in explainable AI applied to biomechanics — the research side of a technology career, and where my instincts about trustworthy AI features come from.
01 Context
Alongside my day-to-day work as a software engineer, I've been a researcher with the UNLOC (Unlimited Locomotion) group at Universidad San Jorge since 2024, and a PhD candidate in Health Sciences since 2025. The research develops explainable AI and leakage-free machine-learning pipelines for clinical gait and foot-pressure data — and it shapes how I integrate models into the products I build: if a system can't explain itself, it doesn't ship.
ORCID 0009-0004-7663-288802 Doctoral research
The thesis, in plain terms
Teach a model to recognise how people walk, then force it to justify every call it makes.
2025 — present PhD in Health Sciences
Explainable AI for the biomechanics of human locomotion
Segmented normality patterns and biomechanical pre-diagnosis.
03 Method
Six steps, in order
Select a step to see what happens in it.
01
Clinical problem definition
Frame locomotion pathologies where explainability is critical. Work with the UNLOC group and clinicians to define segmented normality and pre-diagnosis targets.
02
Data acquisition
Clinical gait and foot-pressure data — baropodometry, optical motion capture, OptoGait and inertial sensors — across diverse cohorts.
03
Leakage-free ML pipelines
Machine-learning pipelines with rigorous, leakage-free validation, so reported performance reflects genuine generalisation rather than data leakage.
04
Explainable AI
SHAP, LIME, Grad-CAM and counterfactual explanations, applied to make model decisions transparent and clinically meaningful.
05
Per-pathology validation
Per-pathology binary classification and segmented normality patterns, assessed against clinical criteria for biomechanical pre-diagnosis.
06
Dissemination
Systematic review (PRISMA; PROSPERO) and methodological manuscripts sharing reproducible, explainable methods with the research community.
04 Focus areas
Research domains
Locomotion biomechanics
Analysing clinical gait and foot-pressure data — baropodometry, optical motion capture, OptoGait, inertial sensors — to characterise segmented normality patterns.
Leakage-free ML
Reproducible pipelines and per-pathology binary classification, so reported performance reflects true generalisation rather than optimistic leakage.
Clinical decision support
AI that enhances rather than replaces clinical judgement — explanations designed around clinician and sports-science workflows, validated against clinical standards.
05 Insights
What the work has taught me
- 01
Explainability increases adoption
Clinicians and sports-science professionals are significantly more likely to act on AI recommendations when they understand the reasoning behind them.
- 02
Leakage is the silent killer
Much of the reported accuracy in locomotion ML collapses under leakage-free validation. Honest pipelines matter more than headline metrics.
- 03
One technique isn’t enough
SHAP, LIME, Grad-CAM and counterfactuals each reveal a different facet of a decision. Combining them gives a fuller, more trustworthy explanation.
- 04
Normality is segmented
There is no single "normal" gait. Segmenting normality patterns is what makes biomechanical pre-diagnosis meaningful and fair.
06 Publications
Work in progress
Systematic review
Explainable AI in Human Locomotion Biomechanics: A Systematic Review
A PRISMA-guided systematic review (registered on PROSPERO) mapping how explainable AI is applied across human-locomotion biomechanics.
PRISMAPROSPEROXAIBiomechanics
In preparation Preprint — coming soon
Methodological manuscript
Per-Pathology Binary Classification with SHAP-Based Explainability
A methodological manuscript on leakage-free per-pathology binary classification of locomotion data, with SHAP-based explainability of model decisions.
SHAPClassificationGait
In preparation Preprint — coming soon
07 Academic appointments
Teaching and academic leadership
University Lecturer
Universidad San Jorge
Teaching Artificial Intelligence, Cybersecurity and Bioinformatics.
Director, MSc in Advanced Technologies for Mobile Devices
Universidad San Jorge
Academic direction of the master’s programme.
08 From lab to field
Where research meets sport
My doctoral research on explainable AI and human locomotion has a natural applied home: sport. Through my innovation work with national and regional judo federations, the same methods — leakage-free pipelines, interpretable models, biomechanical pre-diagnosis — reach injury prevention, athlete monitoring, and the clinicians and coaches who support high performance.
The goal is the same in the lab and on the field: AI that the people behind human performance — sports scientists, physiotherapists, and the athletes themselves — can genuinely understand, question and act on.
Collaboration
Research partnerships?
Open to collaborating on explainable AI, locomotion biomechanics and applied research. Full academic record on ORCID.