Core Engineering Team

Built a Qualcomm Hexagon NPU-optimized LLM inference engine for SA8295P embedded device with mixed-precision quantization, speculative decoding, chunked loading and custom profiling support, empowering efficient deployment across NIO's automotive platforms.

Designed and deployed a psychologically reflective hybrid (rule-based + LLM) conversational agent to facilitate health behavior change at NUS AI4SG Lab, and analyzed HCI trade-offs across dialogue system frameworks.

Implemented a protein multi-structure alignment tool using GNN-based 3D representation learning, enhanced with Geometric Vector Perceptrons and contrastive training on Foldseek-derived data, enabling applications in protein information retrieval.