Selected Projects & Systems
A curated collection of production-grade Machine Learning, Deep Learning, and MLOps systems built with performance, modularity, and reproducible research in mind.
-
Multimodal Fake News Detection via BERT & Vision Transformers 96.66% SOTA Acc
End-to-end framework integrating text encodings from DeBERTa-v3 with visual patch embeddings from Swin Transformer and CLIP-ViT. Features cross-modal attention, gated linear units, and 5-fold cross-validation delivering 96.75% Macro F1.
PyTorch Hugging Face Swin Transformer DeBERTa CLIP-ViT -
End-to-End MLOps Disease Classification Pipeline 96.55% CV Acc
Production-ready automated image classification system for poultry disease diagnostics. Features dataset versioning with DVC, experiment logging with MLflow, containerization with Docker, and automated deployment to AWS EC2 via GitHub Actions.
DVC MLflow Docker FastAPI AWS EC2 GitHub Actions -
Real-Time Customer Churn & Score Prediction Engine <50ms Latency
High-throughput ML inference engine benchmarking 8 ensemble models (XGBoost, CatBoost, GradientBoost). Provides local and global model explainability via SHAP, RESTful microservices via FastAPI, and interactive dashboards via Streamlit.
XGBoost CatBoost FastAPI SHAP Streamlit Docker -
AllAgeCalculators.com Live Web Platform
High-precision chronological analytics platform delivering microsecond client calculations, interactive visualization widgets, responsive UI glassmorphism, and comprehensive SEO optimization.
JavaScript ES6+ Chronological Algorithms SEO Optimization
