What we're building right now.

Cache-inspired federated learning framework for efficient communication and scalable edge intelligence.

End-to-end autonomous driving stack built on Mamba state-space models for efficient long-horizon perception and planning.

Adaptive selective aggregation strategy for federated learning across heterogeneous model architectures. Tier-based partial backbone sharing with adaptive blending and validation-based rollback to mitigate negative transfer — evaluated on a fleet of seven autonomous driving models in CARLA.

Quantum qubit simulator for quantum computing research and experimentation. Designed for algorithm prototyping and hybrid quantum-classical workloads.

A new model architecture for time-series forecasting, exploring novel designs for sequential data prediction with a focus on accuracy and efficiency.