Oguz Bektas

Oguz Bektas

Data Scientist & Researcher · ML, Deep Learning, LLMs

Industrial AI · Predictive maintenance · Explainable ML

About

Data scientist building machine learning systems for industrial applications. I work across the full stack — from raw sensor signals and time-series feature engineering to deep learning models, LLM-powered tooling, and explainable AI for high-stakes decisions.

My focus is making complex models trustworthy and useful in production: predicting failures before they happen, estimating remaining useful life, and turning black-box predictions into something engineers can actually act on.

Focus Areas

Predictive Maintenance & PHM Remaining useful life estimation, condition monitoring, fault diagnosis, and degradation modeling for industrial assets.
Predictive maintenance

Predictive Maintenance & PHM

Building data-driven systems that watch industrial assets in real time — catching failures before they happen, estimating remaining useful life on , and turning raw sensor noise into actionable health indicators for maintenance teams.

Machine Learning & Deep Learning Time-series models, transfer learning, foundation models, and production ML pipelines.
Machine learning and deep learning

Machine Learning & Deep Learning

End-to-end ML systems for sequential industrial data — from feature engineering on raw signals to fine-tuning time-series foundation models on domain-specific corpora, with production deployment across modern data-platform stacks.

Large Language Models LLM engineering, retrieval-augmented generation, prompt design, and AI-powered workflows.
Large language models

Large Language Models

Applied LLM engineering — retrieval pipelines with vector search and hybrid retrieval, structured generation with schema validation, multi-agent orchestration with tool use, and evaluation harnesses that hold up when language models do real work.

Explainable AI SHAP, feature attribution, model interpretability, and decision-support systems for engineers.
Explainable AI

Explainable AI

Making model outputs interpretable for the engineers who act on them — SHAP-based feature attribution, counterfactual explanations, CAM-guided subsequence selection for multivariate time series, and decision-support dashboards that surface the why behind each prediction.

Stack

Python PyTorch TensorFlow scikit-learn PySpark Databricks MLflow Delta Lake SQL Azure LLMs RAG SHAP Time Series Signal Processing

Currently

Working on industrial AI projects — prognostics, LLM-powered tooling, and applied research with a clear path to production.

Connect

LinkedIn →