Gokul
K.C.

Cloud platform enabling developers to test IoT sensor logic without physical hardware by executing code inside isolated Docker containers. Architected a Redis Pub/Sub message bus to stream real-time execution logs from multi-worker Uvicorn backend to WebSocket clients with sub-second latency. Resolved async event-loop contention in concurrent log streaming via a shared Redis state architecture. Features live digital twin sensor state updates, configurable CPU/memory sandboxes, and execution history.

3-agent cooperative reinforcement learning system where autonomous drones track an intruder in a 20×20×10m PyBullet-simulated airspace using a PettingZoo parallel MARL environment. Designed an EMA-based trust scoring mechanism with adversarial communication modeling (Bernoulli packet drops + Gaussian spoofing) to maintain reliable coordination under comms failure. Achieved >60% capture success rate under 30% simulated communication loss — outperforming a baseline MARL system that drops below 40%. Full MAPPO training pipeline with WandB experiment tracking and ONNX export.

Edge-cloud smart agriculture platform using ESP32 + TinyML + LoRa + Raspberry Pi for real-time farm condition monitoring. Achieves over 90% on-device classification accuracy with sub-50ms inference latency in low-connectivity environments. Integrates IoT sensor data, weather APIs, market price feeds, and multilingual AI advisory (English, Hindi, Kannada, Tamil). Reduces water wastage by 35% through data-driven automated irrigation strategies.

Hybrid file integrity monitor combining SHA-256 cryptographic hashing with behavioral anomaly detection. Detects ransomware-like patterns in real-time via filesystem event monitoring — rate-of-change analysis, Shannon entropy spikes, extension patterns, and directory traversal detection. Features automated policy-driven rollback. Tested across 500+ simulated attack scenarios.