Ashak Umesh

Ashak Umesh — ML researcher & full-stack developer | HITL-RL · Deepfake Detection · LLM Infrastructure · Web

Full-stack quiz platform (Django + JS) — B.Tech team mini-project. Quiz creation, auto-grading, result tracking.

Human-in-the-Loop TD3 reinforcement learning for autonomous driving in CARLA. Multi-hazard environment (5 concurrent obstacle types), real-time human intervention via keyboard/mouse.

Cross-plane closed-loop defense against protocol-compliant DoS attacks on TSN-based Automotive Ethernet. Detects timing/rate-evasion attacks that respect IEEE 802.1Qci PSFP filters, then adapts PSFP via Q-learning RL — while formally preserving ASIL-D streams. M.Tech thesis · NIT Calicut · OMNeT++ · INET · scikit-learn.

No project description available.

Video-based deepfake detection using ResNeXt + LSTM (spatial + temporal). Trained from scratch on FaceForensics++, 94% test accuracy. Django web app for real-time inference.