Experience

Software Engineering Intern May 2026 – Aug 2026
Solos Technology · Hong Kong Science Park
  • Conducted systematic benchmarking and profiling of sub-billion parameter lightweight LLMs/SLMs under strict smart glasses edge constraints, evaluating memory footprints, inference latency, and schema compliance to establish deployment trade-offs.
  • Integrated PaddleOCR V5 into an automated business card extraction workflow, engineering async polling and schema-guided prompt pipelines to enhance OCR reliability in low-contrast, CJK-heavy scenarios.
  • Built analytical pipelines for wearable device telemetries, isolating critical hardware failure modes to guide cross-functional firmware and hardware reliability enhancements.
Open-Source Contributor Feb 2026 – Present
Intel OpenVINO Toolkit
  • Landed upstream PR #34577 (merged Oct 2026) in the OpenVINO PyTorch frontend, adding conversion support for aten::_fake_quantize_learnable_per_channel_affine across the TorchScript and FX / torch.export paths — unblocking QAT models with learnable quantization parameters (resolves issue #29682).
  • Diagnosed low-level performance bottlenecks in the asynchronous execution pipeline, focusing on hardware-specific routing for edge inference targets.
Research Assistant — 3D Point Cloud Optimization Mar 2025 – Sep 2025
Autonomous Driving Perception Lab
  • Investigated point cloud transmission and compression strategies for autonomous driving perception pipelines.
  • Extended PointNet++ with a binary segmentation head for foreground/background classification, trained and evaluated on KITTI and V2X datasets.