São Paulo, Brazil 🇧🇷 | Available for Remote Global Roles (US / EU / Worldwide) & Hybrid/On-site in SP
I am a Senior Machine Learning Engineer and Data Scientist with an uncommon hybrid background: a decade of hands-on mission-critical experience in hospital biomedical engineering (calibrating Computed Tomography scanners and life-support units) combined with production-grade AI and distributed software engineering.
This gives me an irreplaceable edge in mission-critical environments: I understand the physical nature of sensor noise, raw DICOM radiography, and healthcare workflows from the hospital floor to the server rack.
- 🩺 HealthTech & Level 4 Interoperability: Production implementation of the international openEHR standard (Archetypes & Compositions) and HL7 FHIR R4, architected under strict Privacy by Design (LGPD/HIPAA).
- 🔬 Math & Algorithms From Scratch: Creator of the proprietary CFIS (Class-Feature Interaction Selection) vector algorithm in pure NumPy broadcasting.
- 🩻 Computer Vision & Explainable AI (XAI): ResNet50V2 Transfer Learning paired with Grad-CAM thermal maps (
tf.GradientTape) and volumetric 3D CT segmentation with PyTorch U-Net (0.89 Dice). - 🌊 Big Data & Distributed Lakehouses: PySpark 3.5 & Delta Lake Medallion Lakehouses with ACID transactions, time-travel, and scalable feature stores on MIMIC-IV ICU data.
- ⚙️ End-to-End MLOps & Governance: MLflow Model Registry gates, continuous Evidently AI data/concept drift monitoring, Docker multi-stage builds, and automated GitHub Actions CI/CD with PyTest regression testing.
- ⚖️ Statistical Rigor & Causal Inference: Clinical A/B testing with False Discovery Rate (FDR) control and Causal Inference via DoWhy & Propensity Score Matching (PSM).
┌─────────────────────────────────────────────────────────────────────────────────────────┐
│ SADC-CKD: CLINICAL DECISION SUPPORT MICROSERVICES │
│ │
│ ┌─────────────────────┐ HTTP / JSON ┌────────────────────────────────┐ │
│ │ sadc-frontend │ ───────────────────────> │ sadc-api │ │
│ │ (React 18 + Vite) │ <─────────────────────── │ (FastAPI Assíncrono) │ │
│ │ Served by Nginx │ └──────────────┬─────────────────┘ │
│ └─────────────────────┘ │ │
│ │ │
│ ┌───────────────────────────────────────────────────┴─────────────┐ │
│ │ │ │
│ ▼ ▼ │
│ ┌────────────────────────┐ ┌─────────────────────┐ │
│ │ Master Patient Index │ │ Clinical Store │ │
│ │ PostgreSQL 16 Alpine │ │ MongoDB 7.0 │ │
│ │ (PII / Identificável) │ │ (openEHR Archetypes)│ │
│ └────────────────────────┘ │ Linked only by UUID │ │
│ └─────────────────────┘ │
│ ▲ ▲ │
│ └───────────── PRIVACY BY DESIGN (LGPD / HIPAA) ──────────────────┘ │
│ Physical Database Segregation │
└─────────────────────────────────────────────────────────────────────────────────────────┘
| Repository | Technology Stack | Key Metrics & Highlights |
|---|---|---|
| CKD_OPENEHR | Docker Compose, FastAPI, React 18, Nginx, MongoDB, Postgres | 99.60% Recall, openEHR Level 4 Compositions, LGPD PII segregation. |
| openehr-fhir-clinical-converter | Python 3.10+, Pydantic V2, openEHR RM, HL7 FHIR R4 | Bidirectional translation engine with <5ms latency per clinical transaction. |
| CKD-DL-CRISPDM | TensorFlow/Keras, ResNet50V2, Grad-CAM, Gradio, OpenCV | Explainable AI via Grad-CAM thermal maps eliminating medical black-box barriers. |
| dicom-segmentation-unet-kidney | PyTorch, U-Net, pydicom, Hounsfield Units (HU) | Volumetric 3D CT segmentation reaching 0.89 Dice Similarity. |
| biomedical-signal-anomaly-detector | SciPy DSP, Pan-Tompkins, PyTorch Autoencoders | 98.7% sensitivity in detecting real-time cardiac arrhythmia in ECG telemetry. |
| Repository | Technology Stack | Key Metrics & Highlights |
|---|---|---|
| mlflow-model-registry-pipeline | MLflow, Scikit-Learn, Docker, Model Governance | Automated staging-to-production promotion gates with complete artifact lineage. |
| data-drift-detector-evidently | Evidently AI, Kolmogorov-Smirnov, PSI, Scikit-Learn | Continuous statistical drift monitoring triggering proactive retraining alerts. |
| automated-ci-cd-model-deploy | GitHub Actions, Docker multi-stage, FastAPI, PyTest | Zero-downtime automated deployment with automated regression test gates. |
| iris-classification-en | Python 3.12, Docker Compose, XGBoost, PyTest | Production-ready CRISP-DM pipeline with 100% automated test coverage. |
| Repository | Technology Stack | Key Metrics & Highlights |
|---|---|---|
| pyspark-healthcare-lakehouse | PySpark 3.5, Delta Lake 3.1, ACID, Partitioned Parquet | Medallion architecture (Bronze/Silver/Gold) with automated PII masking. |
| spark-feature-store-mimic | PySpark, Delta Lake, Window Functions, Time-Travel | Distributed ICU feature store preventing 100% of pre-inference data leakage. |
| azure-databricks-churn-etl | Azure Databricks, PySpark, Delta Lake, Cloud ETL | Scalable telemetry pipeline optimizing cloud costs through smart partitioning. |
| Repository | Technology Stack | Key Metrics & Highlights |
|---|---|---|
| clinical-rag-medical-guidelines | ChromaDB, LangChain, Embeddings, FastAPI | Clinical RAG with cosine similarity and mandatory source citations (zero hallucination). |
| medical-records-summarizer-llm | Transformers, SpaCy Clinical, Pydantic, RegEx Sanitizer | Automated SOAP note summarization with 100% PII de-identification before inference. |
| multimodal-medical-report-generator | PyTorch, Vision Transformers (ViT), Multimodal LLM | Multimodal feature fusion aligning medical radiography scans with clinical histories. |
| Repository | Technology Stack | Key Metrics & Highlights |
|---|---|---|
| clinical-ab-testing-framework | Statsmodels, SciPy, FDR Benjamini-Hochberg | Statistical power sizing (MDE), non-parametric hypothesis tests, and multiple testing correction. |
| causal-inference-healthcare-outcomes | DoWhy, Propensity Score Matching (PSM), DAGs | Average Treatment Effect (ATE) estimation validated via placebo refutation tests. |
| multi-armed-bandit-triage | NumPy, Thompson Sampling, UCB1, Epsilon-Greedy | 42% cumulative regret reduction compared to static traditional A/B testing. |
| equipment-predictive-maintenance-roi | Lifelines, Weibull Survival Analysis, Monte Carlo | Remaining Useful Life (RUL) modeling showing up to 38% reduction in downtime costs. |
| patient-churn-and-no-show-analyzer | XGBoost, SHAP Values, CRISP-DM, FastAPI | 0.88 AUC-ROC identifying key behavioral churn drivers and recovering idle revenue. |
| Repository | Technology Stack | Key Metrics & Highlights |
|---|---|---|
| streamlit-cfis-ckd-dashboard | NumPy Broadcasting, Streamlit, SciPy, Jinja2 | Proprietary CFIS algorithm benchmarking univariant ANOVA vs Gini importance. |
| DEEP_LEARNING_TITANIC | PyTorch (nn.Module), GPU/CUDA, DataLoader |
Object-oriented deep neural network with custom mini-batches and Dropout regularizers. |
| PLN-FERNANDO-PESSOA | Keras/TensorFlow, TF-IDF, NLTK, BeautifulSoup4 | 80.0% accuracy on 4 literary heteronyms with confidence threshold rejection. |
| Thyroid-Cancer | Jupyter, Pandas, Scikit-Learn, CRISP-DM | Bilingual clinical predictive modeling and EDA published in Portuguese & German. |
| Domain | Technologies |
|---|---|
| Languages |     |
| Deep Learning & Vision |     |
| Machine Learning & Stats |    |
| Big Data & Lakehouse |    |
| MLOps & DevOps |     |
| Databases |    |
| Healthcare Standards |     |
| GenAI & NLP |    |
Grandes grupos hospitalares e empresas de dados (Einstein, Sírio, Fleury, Dasa, Hapvida, seguradoras e consultorias) frequentemente sofrem com cientistas de dados que desconhecem a rotina clínica e a regulamentação sanitária.
- Eu trago mais de 10 anos de experiência real em ambientes hospitalares críticos resolvendo problemas em salas de emergência e centros cirúrgicos.
- Conheço a física do equipamento médico, o fluxo de prontuários eletrônicos e a LGPD na prática, implementando microsserviços com segregação física de PII.
Global hiring managers need autonomous, mature engineers who write reliable, maintainable software and communicate transparently.
- Production-Grade Engineering: Every single project features modular clean code, strict Pydantic schemas, PyTest automated testing suites, Docker containerization, and automated CI/CD pipelines.
- International Standards Leadership: Mastery in openEHR and HL7 FHIR R4, which are the required backbone for European and US digital health networks.
- Multilingual Capability: Fluent in English for technical architecture and code documentation, with conversational technical German (
thyroid_eda_portfolio_deutsch_val.ipynb).
I am actively open to discussions regarding Senior Data Scientist, Machine Learning Engineer, and Healthcare AI Architect roles.
- 📧 Email: pauloelias41@gmail.com
- 💼 LinkedIn: linkedin.com/in/paulo-generozo
- 🌐 GitHub Repositories: github.com/generozo-paulo
- 🔬 R&D Algorithms: github.com/paulo-generozo