Senior Data Scientist · Londrina, Brazil

Production ML
for industrial
domains.

I build end-to-end machine learning systems for industrial domains, where models meet sensors, pipelines, and high-stakes decisions. Five years shipping production ML across oil & gas, finance, and automotive — with depth in predictive maintenance, anomaly detection, and MLOps platforms.

Engineer first, data scientist by craft.

I started in electrical engineering, took a master's in electronic systems where I published a comparative study on photovoltaic power forecasting, and then crossed into ML for crash-safety research at Technische Hochschule Ingolstadt as a DAAD scholarship recipient. That academic detour set the tone for the rest of my work: physics-aware modeling, careful evaluation, and a low tolerance for results that don't survive contact with reality.

Today I'm a Senior Data Scientist in oil & gas, leading technical work on a fuzzy inference system for offshore process safety in partnership with a major energy operator and a top research university. Before that, I built corrosion-detection computer vision, anomaly-detection pipelines on offshore sensor networks, and predictive-maintenance models for industrial spindles across dozens of assets.

I'm equally at home shipping the model and shipping the infrastructure around it. I designed my team's deployment framework, QA library, and automated backtesting pipeline — the unglamorous plumbing that turns prototypes into 100+ production models with auditable behavior. $1M+ in client savings and contributions to strategic enterprise accounts followed.

I'm currently open to Senior Data Scientist and Senior MLE roles in the US and EU, including fully remote. If your team works on industrial AI, safety-critical ML, or production ML platforms, I'd love to talk.

A few projects worth talking about.

01 · Process Safety · Oil & Gas 2025 — Present

Fuzzy BowTie inference system for offshore process safety

Lead technical owner of a risk-assessment system combining fuzzy inference with probability trees, delivered in partnership with a major energy operator and a top research university. Designed the architecture, ran end-to-end data analysis on highly fragmented industrial data, and built a tag-clustering algorithm based on business rules that drastically reduced base complexity. First production milestone delivered with full stakeholder approval.

~90%data complexity reduced
2external partners led
Fuzzy InferenceProbability TreesPythonProcess SafetyStakeholder Mgmt
02 · MLOps · Platform 2025

Accelerated deployment framework & QA library

Architected the team's end-to-end deployment workflow and a modular Python QA library for unit and integration testing across dev/prod environments. Restructured the model catalog (50+ legacy entries) and authored the team-wide development & deployment guidebook. Adopted across the data science team and integrated into the lifecycle of 100+ production models.

13×deploy throughput
96%validation cycle cut
100+production models
MLOpsPythonQA FrameworksAzure DevOpsDocumentation
03 · Computer Vision · Industrial Inspection 2024 — 2025

Corrosion detection on offshore equipment

Led a computer vision PoC for corrosion detection on offshore industrial installations using DeepLab semantic segmentation, with data augmentation and patching techniques for high-resolution 360° imagery. Built and supervised the labeling pipeline with LabelMe and established baselines for production rollout.

~70%F1 / IoU
1,500images labeled
PyTorchDeepLabSemantic SegmentationLabelMeData Augmentation
04 · Predictive Maintenance · Industrial 2024 — 2025

Predictive maintenance & sensor anomaly detection

Built predictive-maintenance models for industrial spindle telemetry across 30+ assets and 4 clients, plus anomaly-detection pipelines for offshore sensor data. Early identification of an asset failure was instrumental in a key client renewing a 1-year contract. Also developed a knowledge-based sensor-comparison model with adaptive normalization to detect faults across redundant and analog sensors at vessel scale.

30+industrial assets
1,000+sensors per vessel
Time-seriesAnomaly DetectionSensor AnalyticsPythonDatabricks
05 · Probabilistic ML · Public Utilities 2022 — 2024

Damage prediction for a metropolitan gas network

Designed a probabilistic ML model to predict damage risk to São Paulo's underground gas distribution network from third-party construction work (20–30 sites/day). Output directly drove field-engineering routing and prioritization. Also delivered a billing-rule optimization model and a People Analytics solution that identified improper reimbursements.

~$55Kmonthly cost savings (USD)
+70%field productivity
~$165Kadditional savings (USD)
Probabilistic ModelingPythonAzure MLDatabricksh2o.ai
06 · Research · Surrogate Modeling 2020 — 2021 · DAAD · Germany

Gaussian Process surrogates for crash simulation

Developed Gaussian Process surrogate models at CARISSMA / Technische Hochschule Ingolstadt (DAAD scholarship) to replace computationally expensive Finite Element crash simulations, while maintaining low Mean Absolute Error against the FE ground truth. Published in VDI on metamodel robustness and predictive power considering human diversity.

24×simulation speedup
72h → 3hruntime reduced
Gaussian ProcessesSurrogate ModelingFE SimulationPythonResearch

Peer-reviewed.

VDI · 2022 · pp. 313–326
Rapid Estimation of Occupant Crash Behavior Considering Human Diversity: Robustness, Data Intensity and Predictive Power of Metamodels
F. Plaschkies, O. Vaculin, A. Pelisson
doi.org/10.51202/9783181023877-313 →
ENIAC · 2020 · pp. 555–566
Comparative Study of Photovoltaic Power Forecasting Methods
A. Pelisson, T. Covoes, A. Spengler, P. Jaskowiak
doi.org/10.5753/eniac.2020.12159 →

Tools of the trade.

Languages
Python · SQL · PySpark · Bash
ML / DL
PyTorch · TensorFlow · scikit-learn · XGBoost · DeepLab · semantic segmentation · anomaly detection · time-series · Gaussian Processes · fuzzy inference · Bayesian methods
MLOps
MLflow · Docker · Git · Azure DevOps · CI/CD · model monitoring · QA frameworks · automated backtesting
Cloud & Data
Azure (ML, Data Factory, Synapse, Pipelines, Storage) · Databricks (Lakehouse, Delta Lake) · AWS SageMaker · h2o.ai
Domains
Computer Vision · Predictive Maintenance · Process Safety & Risk · Anomaly Detection · Credit Risk · Sensor Analytics
Languages, human
Portuguese (native) · English (C1) · German (A2) · Spanish (A2)

Hiring, collaborating, or just curious — drop a line.