Turgun Kashgari is a Senior Associate Researcher in the Media Analytics Department at NEC Laboratories America. His work focuses on efficient deployment and optimization of deep learning models, large language models, and vision-language models on local GPUs, CPUs, and edge devices. He earned his M.S. in Computer Science from the University of Central Florida in 2017, specializing in computer vision, machine learning, and data mining, and his M.S. in Communication and Information Systems from Shanghai Jiao Tong University.
Over the past several years, Turgun has worked extensively on real-time AI deployment, model quantization, model optimization, knowledge distillation, and low-latency inference. His recent work focuses on LLM and VLM optimization, including weight quantization, KV-cache quantization and optimization, speculative decoding, disaggregated serving, vocabulary reduction, and efficient serving architectures. He has hands-on experience with major deployment frameworks such as TensorRT-LLM, vLLM, LMDeploy, llama.cpp, OpenVINO, and SGLang.
Turgun has also contributed significantly to autonomous driving projects, deploying perception models and LLM/VLM-based planning models directly on vehicle platforms. He is experienced with ROS2, Autoware Universe, real-time perception pipelines, edge AI systems, and scalable model-serving infrastructure. His work supports NEC Labs’ efforts to build reliable, efficient, and high-performance AI systems for autonomous driving, intelligent transportation, smart cities, enterprise automation, and other real-world applications.
Driving vision-language models (VLMs) must accurately understand scenes across diverse conditions defined by Operational Design Domains (ODDs), yet verificationremains sparse: many slices are missing, making empirical failure rates unreliable. We propose SLICESCORER, a deterministic scoring rule for missing-slicerecommendation that combines (i) an exposure-based coverage prior to prioritize rare, under-tested regions, and (ii) a neighbor-failure prior that propagatesrisk from similar tested conditions. SLICESCORER is deliberately simple- interpretable, auditable, and conservative- properties essential for safety-criticalvalidation. For stress testing beyond the declared ODD, we embed SLICESCORER within SLICENAV, an LLM-orchestrated verification pipeline where the modelinterprets developer queries to select relevant operators (triage, scoring, acquisition, evaluation) and vocabulary extensions, composing verification workflows whilekeeping all scoring deterministic and auditable. Experiments on three driving VLMs(WiseAD, DriveMM, Cosmos-Reason2-2B) show that SLICENAV surfaceshigh-risk coverage gaps more effectively than prior slice-discovery methods while maintaining diverse recommendations across the condition space. Ablations confirm both scoring components contribute, and qualitative analysis demonstrates end-to-end workflows from developer query to targeted evaluation.
