Chinese Journal of Physics · Elsevier · Volume 103, pp. 1881–1895 · 2026
Role: First Author
Co-authors: Muhammad Saeed, Fakhrud Din, Farhat Ullah, Sami Ullah
A physics-informed machine-learning workflow for large-scale screening and validation of MOF candidates for Type-II heterojunction photocatalysis with g-C₃N₄.

20,152 MOFs → 983 high-confidence candidates
ROC-AUC 0.912 · Top-50 Precision 98% · External Family-Level Recall 100% · Precision 85.7% (closed-shell safe-label subset, 11 families) · +27.6 pp over physics-rule baselines · ~20.5× estimated screening-cost reduction

Methods   Random Forest · XGBoost · SHAP · Physics-Informed Descriptors · Butler–Mulliken Band Alignment · Physical Constraints

Interpretable drivers:   Metal ionic radius · linker π-conjugation · structural stability · pore accessibility

Source Code    Paper

A reliability-aware SciML platform for composition-based thermodynamic stability screening of inorganic crystals.

Pipeline   132 Magpie Descriptors → XGBoost → Calibration → Uncertainty → OOD Risk → Candidate Ranking

Benchmark   ROC-AUC 0.8521 · F1 0.7082 · AUPR 0.7872 · ECE 0.0246

Stack   Python · XGBoost · FastAPI · Streamlit · SQLite

Repository

Reliability-aware clinical AI combining digital pathology and genomic transcriptomics for colorectal cancer prognosis.

Independent Modality Branches · Transparent Late Fusion · Uncertainty Estimation · Modality Disagreement Detection · Missing-Modality Handling

Stack: Next.js · FastAPI · SQLAlchemy · Docker

Repository

An integrated AI platform spanning battery degradation, energy forecasting, photovoltaic modeling, digital twins, and constrained dispatch.

Methods   Physics-Regularized Neural Networks · Walk-Forward Validation · P10/P50/P90 Forecasting · HiGHS Optimization

Repository

Independent reproduction of a published inorganic-crystal stability benchmark.

85,014 compounds

ROC-AUC 0.8859 vs 0.8860 reported
F1 0.7550 vs 0.7520 reported

Repository

Cross-domain empirical research spanning superconductivity, battery SOH, cancer prognosis, steel-defect segmentation, and ML pipeline leakage analysis.

Kaggle Notebooks

383,338 surveillance records across 102 countries and territories
Curated for epidemiological modeling, temporal forecasting, and spatiotemporal outbreak analysis.

Explore Dataset

100 stability-ranked genes across 5 cancer cohorts — BRCA, COAD, KIRC, LUAD, PRAD
Designed for biomarker discovery, feature-selection stability, and interpretable oncology ML.

Explore Dataset

Ground-measured solar and meteorological data from 9 weather stations
Supports GHI/DNI/DHI irradiance forecasting and renewable-resource characterization.

Explore Dataset

900 curated Metal–Organic Frameworks with electronic, pore, and geometric metadata
Built for materials informatics, PBE band gap prediction, and small-data representation learning.

Explore Dataset

Complexity vs. utility
TF-IDF + Linear SVM achieved 99.95% accuracy vs 99.53% for DistilBERT.

Distribution shift
Superconductivity: R² = 0.931 on random splits, with +55.5% MAE under chemical-family holdout.

External validation
Multimodal oncology: 0.984 ROC-AUC in cross-validation → 0.681 on external cohorts.

Reproducibility
ECSG reproduction: 0.8859 ROC-AUC vs 0.8860 reported.

Sep 2026 – Present

Alongside teaching and academic responsibilities, I help students develop critical thinking, problem-solving, and professional responsibility, connecting academic learning with practical skills for careers in academia and industry.

Built and deployed production ML systems across predictive lead scoring, computer vision, and LLM-powered automation.

Conducted R&D in interpretability, uncertainty quantification, and reliable ML architectures.

Designed and directed a structured 6-month AI/ML internship program, 9 interns total: 3 completed with distinction - 6 currently active

Public intern testimonial: LinkedIn post

May 2024 – Apr 2026

Computer Science and IT instruction, laboratory teaching, assessment, academic guidance, and student supervision.

Oct 2023 – Dec 2023

Python automation, ML components, data processing, experimentation, integration, and debugging.

Aug 2023 – Oct 2023

Data preparation, baseline modeling, hyperparameter evaluation, and ML integration.

CGPA: 3.81 / 4.00

Thesis: Cancer Biomarkers Identification using Explainable Variational Autoencoders with Ensemble Feature Selection

Research evidence

20,531 → 12 genes · 99.97% accuracy on primary dataset

66,660 → 34 genes · 99.31% accuracy on independent RNA-seq validation

CGPA: 3.77 / 4.00

The Minute-Medical: Multi-Disease Prediction using Machine Learning and Deep Learning