Enterprise banking intelligence system with full-stack Angular + C# + Java architecture and automated PDF/Excel data ingestion via email pipelines.

Designed a hybrid CNN-BiLSTM model with a shared encoder for simultaneous multi-task learning — revenue forecasting, payment behaviour classification (on-time / delayed / early), and anomaly detection on live MySQL transaction data. Incorporated incremental daily learning via a sliding-window technique and Random Search hyperparameter tuning. Delivered a real-time business insights dashboard for merchants and payment gateways. Developed in direct collaboration with Finforz Technologies under the guidance of Dr. K. Selvamani (CEG) and Hari Prasath S (Technical Architect, Finforz).

Built a four-stage IDS pipeline on the CIC-CSE IDS 2018 dataset: (1) Feed-Forward Neural Network for missing-value imputation, (2) CT-GAN for synthetic minority-class data generation (6 attack types, 10k–40k samples each), (3) CNN-BiLSTM for multi-class intrusion classification — achieving ~90% accuracy with near-perfect scores on DoS, DDoS, and brute-force categories, and (4) Fuzzy Logic post-processor to suppress false positives (Confidence Score = 1.00).