Huawei Canada IC Lab

Designed a hash-based lookup system for 4,000+ compiler operators, reducing resolution latency from over 1s to under 20ms.

Developed a custom expression grammar and C++ parser for fusion operators, reducing implementation time by 5x for complex tensor expressions.

Optimized tensor shape propagation speed by 25x in Python/C++ using profiler-guided bottleneck analysis. Reduced 4,000-file operator library load time from 1s to 400ms through data-structure and I/O optimizations.