Jitraj
Jit-fnc
AI & ML interests
I'm building God's Eye, a production-grade hybrid IDS & SIEM platform with a 4-layer ML ensemble: XGBoost (supervised, trained on CICIDS-2018/UNSW-NB15/DARPA/CIC-IoT-2024), a deep Autoencoder trained on benign-only traffic for zero-day detection, a Bidirectional LSTM for multi-step attack chain detection, and Sigma rule integration with CI/CD pipelines. The stack uses calibrated Bayesian ensemble combination, SHAP explainability, and adversarial defenses including stochastic threshold jitter and CUSUM drift accumulation.
I also built IoT Sentinel, an unsupervised Isolation Forest anomaly detection system trained on the CIC-IoT-DiAD-2024 dataset (800K+ network flows across 7 attack categories). It achieves 88.5% accuracy on unseen data with 99.9% attack recall, using 40 engineered CICFlowMeter features, benign-only training, and a 2,000-step threshold sweep. The system runs inference in under 1ms per device.
My experience spans PyTorch (autoencoders, LSTMs, training loops, early stopping, GPU/CPU inference), scikit-learn, XGBoost, SHAP, and large-scale dataset pipelines across CSV, Parquet, and multi-format ingestion with concurrent loading.
Organizations
None yet