Michele E. J. Maestrini BEng
Aerospace predictive maintenance, time-series machine learning, and engineering-grounded analytics
Professional Summary
Data Scientist with an MSc in Data Science and a First Class BEng in Civil Engineering, specialising in aerospace predictive maintenance, time-series analysis, anomaly detection, and physics-aware machine learning. Michele brings a structured engineering mindset to safety-critical problems, with a focus on reliability, traceable evaluation, and practical decision support.
Current flagship work includes FusionCore v0, a physics-aware predictive maintenance pipeline for turbofan Remaining Useful Life estimation on NASA C-MAPSS, combining zero-leakage preprocessing, regime-aware normalisation, physics-informed feature engineering, and maintenance-oriented evaluation. Alongside this, Michele’s portfolio includes a lightweight few-shot learning thesis pipeline and the Moon Landing platform, demonstrating full-stack delivery, database architecture, and AI-assisted product build capability.
Core Competences
This part of Michele’s profile centres on predictive maintenance, anomaly detection, time-series analysis, deep learning, feature engineering, zero-leakage preprocessing, and physics-aware modelling. The underlying stack spans Python, SQL, TensorFlow, scikit-learn, XGBoost, Jupyter, Git, and AI-assisted development workflows.
The detailed breakdown of tools, methods, and competencies now sits on its own page so the main Curriculum Vitae view stays focused on profile, evidence, and career direction.
Education
MSc Data Science
Birkbeck, University of London
Advanced training in machine learning, deep learning, and statistical learning, anchored by a few-shot learning thesis delivering 98.75% accuracy.
BEng Civil Engineering
University of Westminster
Foundation in structural analysis, impact loading analysis, and engineering mathematics, providing the physical intuition behind the applied AI work.
Selected Project Focus
Aerospace Predictive Maintenance & Time-Series RUL Estimation
Designed an end-to-end turbofan Remaining Useful Life pipeline on NASA C-MAPSS FD001–FD004, focused on predictive maintenance rather than generic regression. The work spans 160,359 engine-cycle observations from 709 run-to-failure trajectories and is structured around engineering credibility as much as model accuracy.
- Built zero-leakage preprocessing with grouped engine splits, K-Means regime identification, and regime-specific Z-score normalisation.
- Engineered a 91-feature physics-aware representation including kinematic features, virtual thermodynamic sensors, cumulative fatigue indices, and survival-analysis extensions.
- Achieved RMSE 14.85 cycles, NASA Asymmetric Score 4,336.3, and Critical-band F2 0.9339 on 707 held-out engines.
- Benchmarked XGBoost, TFT, N-HiTS, and DeepAR under literature-standard constraints.
Few-Shot Learning, Transfer Learning, and Optimisation
Developed a lightweight MobileNetV2-based few-shot learning model using Siamese networks, triplet loss, and Bayesian optimisation, delivering 98.75% test accuracy and 98.77% recall for mobile and IoT-suitable deployment.
Full-Stack AI-Orchestrated Portfolio Platform
Designed and built a full-stack web application using Claude Code and Codex on VS Code, combining a multi-agent AI system, interactive chatbot interfaces, a SQLite backend, a document knowledge base, knowledge-graph mapping, and recruiter-facing workflow components.
Professional Experience Highlights
Independent ML Research & Project Development
Focused on FusionCore v0, MSc thesis completion, predictive maintenance, anomaly detection, and applied machine learning research. FusionCore v1 (PiNet, the Predictive In-orbital Network) is under active development.
Career Foundation Across Operations, Logistics, and Client Delivery
Over 20 years of experience across operations management, logistics analytics, client delivery, and business development roles in London and internationally, with recurring strengths in scheduling, process optimisation, structured troubleshooting, stakeholder management, and data-informed decision support.
Additional Strengths
Tooling & Workflow
- Microsoft Office Suite
- AI-assisted development with Cursor AI, Claude Code, and Codex
- Structured research, documentation, and evaluation workflows
- Full-stack build and iteration across content, data, and interface layers
Analytical & Interpersonal
- Critical and abstract thinking
- Complex problem solving
- Attention to detail
- Cross-functional collaboration
- Active listening and communication
