Andrews Cordolino Sobral, Ph.D.

AI Platform Architect | MLOps | AI Orchestration | Ph.D. in Computer Vision & Machine Learning

With 25 years of experience building software systems and over 15 years in AI, machine learning, and computer vision across academia and industry, I specialize in ML platforms, MLOps, and AI orchestration across cloud, HPC, and hybrid environments, helping organizations automate AI pipelines and move AI from research toward production.

AI Team Leadership @ ActiveEon (2017–2025)

At ActiveEon, I led PhD-level AI researchers and architected AI platforms for enterprise and industrial applications, orchestrating and automating research prototypes toward production. My responsibilities included:

• Led PhD-level AI researchers in AI/ML, Data, and Vision Systems
• Architected ProActive AI Orchestration (PAIO) for distributed AI workflows across hybrid cloud and on-premise environments
• Prototyped and deployed LLMs, deep learning, and AutoML pipelines across multi-node, multi-GPU, HPC, and hybrid cloud environments (please see the selected AI use cases below)
• Built custom SDKs and tools for dynamic workflow automation (e.g., ProActive Python SDK, ProActive Jupyter Kernel)
• Developed researcher-facing tooling for hyperparameter optimization, neural architecture search, and evaluation workflows
• Built a centralized MLOps Dashboard for model deployment, monitoring, observability, experiment tracking, and production AI operations
• Led the ExtremeXP (Horizon Europe) runtime implementation for scheduling and executing complex analytics workflows on distributed infrastructure, integrating distributed AutoML, resource monitoring, and dynamic service orchestration.

Selected Enterprise, Industrial & Research AI Use Cases

• (2025) LuxProvide — MeluXina Tier-1 Supercomputer — Designed and implemented reusable, parameterized workflow templates for the MeluXina supercomputer (SLURM-native execution), covering a) multi-node/multi-GPU distributed training with PyTorch DDP, b) AutoML-driven hyperparameter optimization integrated into the DDP training loop, c) MLflow-based experiment tracking with dynamic server provisioning, and d) LLM inference with vLLM across auto-provisioned Ray clusters.
• (2020-2023) IDRIS / Jean Zay (HPE partnership) — Built reusable distributed-training workflow templates for the Jean Zay supercomputer (CNRS/GENCI), enabling multi-node/multi-GPU TensorFlow/Keras training via two orchestration paths: a) SLURM-native execution, and b) Horovod-based workflow for multi-GPU training, both with Singularity/Docker containerization, automated hyperparameter search, and result tracking/visualization via TensorBoard.

• ROTA — Smart Transportation — Built edge AI pipelines for real-time vehicle detection, multi-stream RTSP video analytics, vehicle counting, and smart parking occupancy detection using NVIDIA Jetson, DeepStream, YOLO, and IP cameras, operating continuously (24/7) in real-world production environments.
• IRT Saint-Exupéry / TAS — Contributed to the Reactive & Autonomous Image Chain research project, developing ProActive-orchestrated deep learning workflows for aerial and satellite image analysis, including semantic segmentation pipelines for maritime threat detection and situational awareness.
• SAFRAN Aircraft Engines — Built AutoML-driven predictive maintenance and anomaly detection workflows over aircraft engine sensor data (2,000+ sensors), integrating feature engineering, ML experimentation across multiple algorithms, real-time streaming analysis during test campaigns, and automated engineering report generation.
• Thales Alenia Space (TAS) — Developed industrial AI workflows for satellite manufacturing and telemetry analysis, including deep learning-based wire defect detection on hybrid circuits, sensor anomaly detection using signal processing, TSFresh feature extraction with XGBoost classification, and automated telemetry verification and incident forecasting — reducing per-test analysis time to minutes.
• INRAE / OVHcloud — Led technical workshops and hands-on training for distributed AI and MLOps adoption across cloud and HPC infrastructures, covering deep learning workflows, SLURM-native orchestration, AutoML-driven hyperparameter optimization, and production AI workflow orchestration.
• CHUV / MORPHEMIC Project — Contributed to the design of adaptive AI workflow orchestration and distributed execution architectures for healthcare scientific workloads across hybrid cloud, on-premise, and HPC infrastructures, featuring elastic resource provisioning and utility-driven, accelerator-aware scheduling.

Academic Background & Research

Ph.D. in Computer Vision & Machine Learning
Université de La Rochelle, France (2013-2017)

My research focused on computer vision, optimization, low-rank learning, tensor methods, and real-time video analytics, resulting in:

• 2800+ citations
• h-index: 17
• Publications in IEEE, Elsevier, Springer, CVIU, TIP, TNNLS, ICCV
• Reviewer for international journals and conferences

Open-Source & Contributions

I actively contribute to open-source software for computer vision and machine learning.

🔗 GitHub: github.com/andrewssobral

• BGSLibrary (C++) – Widely used background-subtraction library for moving-object detection.
• LRSLibrary (MATLAB) – Framework for low-rank/sparse decomposition.
• MTT (MATLAB) – Tools for tensor manipulation and decomposition.
• OSTD (MATLAB) – Online stochastic tensor decomposition for multispectral video.
• VDTC (C++ & Python) – Vehicle detection, tracking & counting pipeline.
• GoDec (Python) – Low-rank + sparse decomposition.
• DTT (C++) – Header-only library for seamless data type conversions (Eigen, OpenCV, Armadillo, LibTorch, ArrayFire).

These projects are used by universities, research labs, and industrial organizations worldwide.