AI-Native Software Engineer & Business StrategistBridging Tech, Data & Agentic Automation

Early detection of maternal-infant Low Birth Weight (LBW) risk using complex demographic health survey data (1,863 records) with high-dimensional feature noise.

Engineered a leak-free data preprocessing pipeline and hybrid feature selection reducing 18 maternal-clinical predictors to 12 high-impact features. Benchmarked 14 ML models and constructed a Soft-Voting Ensemble (ExtraTrees + CatBoost) with comprehensive SHAP clinical interpretability.

Severe medical data distribution skew, multi-type tabular features, class imbalance and data leakage risks in predictive obesity classification.

Formulated AIDPCP, an adaptive and distribution-aware automated preprocessing and unsupervised clustering framework combined with Explainable AI modules to deliver transparent decision boundaries.

Early mortality risk stratification in Hepatitis B patients burdened by high-dimensional clinical biomarkers and redundant predictors.

Constructed a novel hybrid feature selection algorithm to isolate dominant prognostic biomarkers and trained gradient boosting ensemble architectures (XGBoost, LightGBM, CatBoost, Random Forest) coupled with SHAP-driven interpretability.

High-fidelity Polycystic Ovary Syndrome (PCOS) diagnosis hindered by high feature dimensionality and opaque 'black-box' medical predictions.

Proposed the AIM-PDCF ensemble architecture combined with QuantumGraphRFE (Quantum Graph-based Recursive Feature Elimination), using graph theory and quantum-inspired ranking to isolate metabolic/hormonal biomarkers with SHAP explanations.

Downsampling resolution loss in standard CNNs when screening for fine-grained pathological textures in chest radiographs.

Architected a dual-stream deep learning framework combining a Global Path (EfficientNet-B0) for macroscopic anatomical context and a Local Path (Shallow CNN) to retain fine-grained textures, integrated via a novel Attention-Gated PCA (AG-PCA) fusion layer.