I built SchedulerAI Orchestrator, an LLM-driven distributed scheduler that cut load imbalance by 49.7% across eight nodes. I implemented consistent hashing, token rings, leader election, and runtime scheduling strategy swaps.
I deployed the system on Render with Docker, FastAPI, WebSocket telemetry, and a React dashboard, while keeping API keys server-side through an LLM proxy.
For SAARTHI, I architected the backend for a team productivity app, including a nine-state task lifecycle, conflict detection, and triage scheduling. I also trained XGBoost and LightGBM stress-scoring models to AUC 0.99 and 0.95.
As a research intern at Thapar Institute of Engineering and Technology, I built a Python pipeline processing 50 GB of satellite imagery daily and shipped an LSTM model for crop-cycle prediction. I also work with federated intrusion detection, sustaining 97%+ accuracy across non-IID client data.
