Real-time hand gesture recognition system interpreting arithmetic operations with both hands. ST-GCN model with 93.65% accuracy and asynchronous strided inference at 30 FPS on CPU.
High-performance algorithmic trading system combining multiple quantitative strategies with a fully vectorised backtesting engine achieving 2,000× acceleration.
Operator-level latency profiling of 24 object detectors on a Tesla T4 with AMP, Conv-BN fusion and TensorRT optimizations reaching up to 5.6× speed-ups. I covered the DETR family and the shared benchmarking tooling.