YWYen-Shi Wang
- Quant employment verification: NOT CONFIRMED. The caller-supplied WorldQuant roles (Quantitative Research Intern 2017-07 to 2017-08; Quantitative Research Consultant 2017-09 to 2018-03) are contradicted by the subject's own CV, which lists a complete, consecutive 2017-2022 timeline with no WorldQuant or any quant/trading/hedge-fund role. His verified employers are NVIDIA (deep-learning inference), Skymizer (DL compiler) and BravoAI (OCR) - none operates as a quantitative investment/trading firm. Identity is supported (Taiwan IMO 2013 silver + CMU + NVIDIA), so the conflict is with the employment claim, not the person.Sep 17, 2026
- Career timeline per CV/site is continuous and consecutive: BravoAI (2018-03 to 2018-09), Skymizer (2019-04 to 2019-07), NVIDIA intern (2020-05 to 2020-08), NVIDIA full-time (2021-03 onward) - leaving no interval matching a 2017-07/2017-08 or 2017-09/2018-03 WorldQuant role.Jul 2022
- Identity confirmed as the Taiwan IMO 2013 contestant: IMO record shows Taiwan, rank 61, silver award, 28 points, and his own CV lists 'Silver Medal, 54th International Mathematical Olympiad (IMO 2013), Santa Marta, Colombia', matching contest/city/year.Jul 2013
- Self-authored CV headline: 'Dedicated C++ programmer with deep learning inference, compilers, performance optimization, and LLM background' - the surface is ML systems/compilers, with no finance, trading or quantitative-investment content anywhere in the CV or personal site.Jul 2022
- Personal site (yen-shi.github.io) confirms NVIDIA promotion to Senior Deep Learning SWE (Data Centers) Sep 2024 and core role in TensorRT for RTX; site tagline 'continual learning as AI does' frames focus on LLM operators, CPU/GPU tuning and compilers.Sep 2024
- Published contact in his CV: email [contact omitted] and phone (+1) 412-218-9816 (Pittsburgh area code, consistent with 2020 CMU enrollment).Jul 2022
- Google Code Jam 2018 Round 1C rank 116 and 2017 ACM-ICPC Asia Hua-Lien silver medal sit between his IMO 2013 silver and his engineering career - a sustained competitive-programming/math profile into university.Apr 2018
Core developer of TensorRT for RTX from project initiation; runtime fusion optimization, trimmed library from 1GB to <200MB, enabled JIT compilation workflow.
Multi-device TensorRT runtime in C++ with NCCL; wrote TensorRT v9.1.0 NeMo demo; FP8 inference support, improved decoding latency 30%.
Automotive Safety; TensorRT Consistency Checker point of contact, deconvolution node validation, ISO 26262 safety verification.
Optimized C++ multithreading server for MLPerf Inference v0.7 BERT; CUDA streams/graphs; +25% throughput.
Open Neural Network Compiler (ONNC); rewrote 21 ONNX optimizations, testing framework, quantization flow for NVDLA.
Optical Character Recognition system with PyTorch/TensorFlow to digitize medical certificate fields; Docker/Flask deployment.
Competition record
Taiwan · IMO