Md. Saidul Islam

Materials scientist (M.Sc., CAU Kiel) fusing first-principles physics with data-driven models across scales.

Currently: a GNoME-inspired structural GNN, live on the Matbench Discovery leaderboard —2nd by F1, 1st by MAE among direct-prediction models, independently verified and merged. Currently building an nequip -type domain-focused interatomic potential and studying sigma phase stress-strain behavior in fe-Cr system via DFT (QE) to upgrade a surrogate model to phase-awareness.

Throughout: models that explain, not just predict. SINDy recovered governing equations from sensor signals (R² ≈ 0.99, ANN-level accuracy, full interpretability);data infrastructures e.g., the public query interface of a neuro-symbolic pipeline, pairing a LoRA-fine-tuned Llama 3.2 with an RDF knowledge graph, deployed on Hugging Face allowing candidate semiconductor screening for functional (e.g., ferroelectric) applications, a DFT → ML → CALPHAD → FEM pipeline extracting, predicting, validating temperature & phase homogenized (298–1200 K, 0–100% Cr), with the TDB-computed Curie/Néel curve overlaid, composition dependent reduce Young's modulus (Eᵣ within ±12% of Hill VRH); parametric MD and DFT studies mapped stiffness and band-gap behavior in Si nanowires and strained graphene.

See key Research Outputs →

Implementation details Git versioned. see GitHub.