Ismail Can Oguz

Open to roles in AI for Science / ML for Materials / Research Engineering

#

I’m a computational materials scientist and scientific ML researcher building reproducible workflows that connect machine learning, atomistic simulation, and density-functional theory. I’m currently a researcher at DIFFER — Dutch Institute for Fundamental Energy Research in Eindhoven, where I work on data-driven materials discovery, foundation interatomic potentials, and multi-fidelity screening for electrocatalysis.

Quick links

#

• CV (PDF): Download
• Google Scholar: https://scholar.google.com/citations?hl=tr&user=VFi3h0sAAAAJ
• ORCID: https://orcid.org/0000-0002-8673-7219
• GitHub: https://github.com/isocan
• Kaggle: https://www.kaggle.com/ismailcanoguz

What I do

#

• Scientific ML for materials — foundation MLIPs, equivariant graph neural networks, high-throughput screening, and model validation against first-principles calculations.
• Atomistic modeling — DFT workflows for surfaces, alloys, adsorption, electrocatalysis, structural stability, and thermodynamic analysis.
• Reproducible research engineering — Python, ASE, pymatgen, automated data pipelines, Colab/GitHub tutorials, and Linux/HPC workflows.
• Applied machine learning — feature engineering, model comparison, interpretability, ensembling, and robust evaluation for scientific and tabular datasets.

Selected scientific ML & data science projects

#

• GNoME, NequIP & Allegro — Au–Pd Formation-Energy Benchmark — Reproducible Colab workflow comparing released GNoME/WBM predictions and open NequIP/Allegro foundation potentials with published PW91-GGA formation enthalpies for ordered Au–Pd alloys.
  Repo: https://github.com/isocan/GNoME-NequIP-Allegro-AuPd-Tutorial

• Energy & Emissions Forecasting — Predicts building energy use and CO₂ emissions from structural attributes, with feature engineering and interpretable regression.
  Repo: https://github.com/isocan/energy-emission-prediction

• Credit Scoring for Limited Histories — End-to-end credit-default modeling with data preparation, model comparison, and SHAP-based interpretation.
  Repo: https://github.com/isocan/credit-scoring-model

• Automated Product Classification (NLP + CV) — Classifies e-commerce products by combining text representations with image models and classical visual features.
  Repo: https://github.com/isocan/automated-product-classification

• E-commerce Customer Segmentation — RFM analysis and clustering for retention, customer-value analysis, and actionable marketing segments.
  Repo: https://github.com/isocan/ecommerce-customer-segmentation

• Fruit Classification — Cloud Pipeline — Distributed preprocessing and scalable inference using AWS, PySpark, EMR, and S3.
  Repo: https://github.com/isocan/fruit-classification-cloud-deployment

• Food Product Analytics for a Nutrition App — Data analysis and scoring for health, environmental, and packaging impacts.
  Repo: https://github.com/isocan/food-product-analysis-nutrition-app

• World Bank EdTech Opportunity Scan — Exploratory analysis of global education indicators to identify promising expansion markets.
  Repo: https://github.com/isocan/world-bank-edtech-opportunities

Latest publication

#

Ismail Can Oguz, Nabil Khossossi, Marco Brunacci, Haldun Bucak, Süleyman Er.
Machine Learning–Accelerated Discovery of Earth-Abundant Bimetallic Electrocatalysts for the Hydrogen Evolution Reaction. ACS Catalysis (2025).
DOI: https://doi.org/10.1021/acscatal.5c04967

Competitions

#

• Kaggle · CAFA 6 Protein Function Prediction — Silver Medal, rank 68 / 2,259 teams.
  Developed a memory-efficient stacking and post-processing workflow that combined complementary public prediction pipelines through score harmonization, agreement-aware logit blending, Gene Ontology propagation, and Information Accretion-aware pruning.
  Reproducible notebook · Competition

• Kaggle · Catechol Benchmark Hackathon (NeurIPS 2025 DnB) — Rank 24 / 227.
  Built a chemistry-aware hybrid ensemble with solvent representations, mixture-symmetry handling, and robust cross-validation.
  Leaderboard

• Kaggle · NeurIPS Open Polymer Prediction 2025 — Team rank 219 / 2,240 — Bronze Medal.
  Competition

Tech I use

#

Python, NumPy, pandas, scikit-learn, PyTorch, JAX, ASE, pymatgen, VASP, FairChem, NequIP, Allegro, MACE, Equiformer/eSEN, PySpark, AWS, Docker, and Linux/HPC.