Joao Santos

Independent R&D – Medical AI Engineer

January 2023 – Present (Remote, Portugal)

Architected and deployed a modality-agnostic deep learning framework for the full spectrum of medical imaging tasks (aligned with MSD benchmarks), bridging raw voxel predictions and standardized clinical analytics (RANO/RECIST).

- Universal Multi-Organ Architecture ("The Tank"): High-performance DynUNet pipeline on NVIDIA RTX A6000, robust to extreme class imbalance (small tumors) and multi-site heterogeneity in CT/MRI.

- Clinical Analytics Engine: Generalized module for RANO (neuro) and RECIST 1.1 (body) metrics, with <3% volumetric deviation.

- 3D Digital Twins: Marching Cubes reconstruction for topologically correct STL meshes.

- Post-processing: Adaptive anatomical refinement to eliminate hallucinations.

- Radiomics: Automatic extraction of morphological/textural biomarkers.

- Ensembles: Meta-ensemble SwinUNETR + CNNs for small-target boost.

- MLOps & XAI: Weights & Biases tracking, Grad-CAM++, error histograms for domain shifts.

ULS Algarve – Clinical Domain Expert | Radiographer

October 2003 – Present (Algarve, Portugal)

+22 years in diagnostic imaging, decision support in emergency/primary care, quality assurance (ALARA protocols), mentoring juniors, and clinical validation of imaging outputs.

Diagnostic Decision Support & Triage: Acting as a critical technical consultant in high-pressure Emergency & Primary Care settings. Frequently consulted by General Practitioners to interpret complex imaging findings, identifying subtle pathologies often missed by non-specialists and mitigating diagnostic errors in real-time.