포지션 상세
This position is with STRADVISION Seoul Office
[About STRADVISION]
We Empower Everything To Perceive Intelligently
With a mission statement of “We Empower Everything To Perceive Intelligently”, STRADVISION is putting all of our effort to make better life for everyone through AI-based camera perception technology. Everyday, we focus on creating AI-based vision perception techonolgy with more than 300 members across 8 offices worldwide and we expect our software to perceive everything precisely & intelligently to make 1% difference in people’s lives. Thus, we are looking for members who would like to join our meaningful journey and face challenges that no one has done it before together at STRADVISION.
[Our Story]
• STRADVISION Seoul Office : https://buly.kr/Edu2Jdr
• STRADVISION’s Core Technology : https://buly.kr/74XS8bi
• STRADVISION's Welfare : https://buly.kr/BpFucdd
[Our Technology]
STRADVISION's Technology : https://stradvision.com/_ENG/dh/technology
• STRADVISION is the FIRST deep-learning based technology start-up company in the world who has obtained ASPICE CL2 certification in 2019.
• STRADVISION has also been honored with the AutoSens Awards for ‘Best-in-Class Software for Perception Systems’(Gold Award Winner) for 2 years in a row(2021, 2022).
• STRADVISION’s outstanding technology was recognized worldwide by successfully completing the Series C funding at KRW 107.6 billion with Aptiv and ZF Group in August, 2022.
• About 167 patents related to autonomous driving/ADAS have been acquired in Korea, Japan, US and Europe. As of today, STRADVISION is actively developing our technology to be differentiated.
• By the first half of 2025, vehicles equipped with SVNet surpassed 4 million units globally, maintaining growth despite economic slowdown and intensifying industry competition.
• In 2025, STRADVISION was ranked 10th in domestic AI competitiveness, following companies such as Samsung, Naver, and LG.
We’re hiring an Embedded AI Engineer to productize MV-Gen2 multi-camera, Transformer-based vision perception models (e.g., object detection / lane / occupancy), and E2E models(e.g., path planner, and control) across multiple automotive SoCs—Renesas, TI (Jacinto/TDA4), Qualcomm, and NVIDIA Orin/Thor.
You will lead INT8 quantization (PTQ/QAT), model porting, runtime/accelerator optimization, and—critically—provide HW-aware redesign guidance to internal DL model teams so architectures remain deployable and efficient under SoC constraints.
This role is a unique opportunity to work on high-impact, cutting-edge research that directly contributes to the development of next-generation autonomous driving systems.
[Key Responsibilities]
The selected candidate will be responsible for designing, developing, and optimizing deep learning models for ADAS/Autonomy.
1) Quantization & Deployment (Multi-Camera Transformers)
• Convert and deploy multi-camera Transformer perception models to embedded inference stacks using PTQ/QAT, calibration pipelines, and accuracy recovery strategies (mixed precision, selective quantization, layer-wise sensitivity).
• Build robust export/packaging flow (e.g., PyTorch ONNX target runtime) and resolve conversion/runtime issues (unsupported ops, precision constraints, graph transforms).
2) HW-Aware Model Redesign Feedback (Internal Technical Leadership)
• Provide actionable feedback to internal DL engineers on SoC-friendly Transformer design :
• Reduce attention/FFN compute & activation memory, manage token/BEV grid size, and avoid/replace ops that break target toolchains.
• Propose architecture patterns that preserve accuracy while meeting embedded constraints (bandwidth, SRAM/DDR pressure, operator support).
• Codify guidelines into “design rules” (allowed/avoid ops, preferred blocks, quantization-robust practices) and drive adoption through reviews and decision logs.
3) Multi-SoC Runtime & Accelerator Optimization
• Optimize latency/throughput/memory for real-time perception across NPU/GPU/DLA/DSP backends :
• Renesas: DRP-AI INT8 PTQ flow (calibration-based static quantization) and translation pipeline integration.
• NVIDIA Orin: TensorRT build/engine tuning, DLA constraints(INT8/FP16) and GPU–DLA workload partitioning.
• Qualcomm: ONNX Runtime QNN Execution Provider/ QNN SDK-based acceleration strategy and deployment validation.
• TI: TIDL quantization modes (PTQ/QAT) and “deployable-by-design” constraints.
4) Tooling & Performance Infrastructure
• Establish repeatable profiling + regression harness for latency/memory/accuracy across SoCs; publish performance dashboards and release artifacts.
• Collaborate with platform/firmware teams to unblock runtime integration and memory/IO bottlenecks.
• MS + 3 years industry or PhD + 1 years (or equivalent), with 3–6+ years hands-on DL in total (vision/ML/robotics).
• Hands-on experience shipping DL models to embedded/edge devices (automotive/robotics/consumer edge), including INT8 quantization (PTQ/QAT) and performance tuning.
• Strong proficiency in Python + C/C++, model debugging, profiling, and deployment automation.
Practical knowledge of at least one deployment stack: TensorRT, ONNX Runtime, QNN, TIDL, DRP-AI, TVM/compilers, etc.
• Experience optimizing Transformer-based vision models (attention bottlenecks, activation memory, mixed precision, operator constraints).
[About STRADVISION]
We Empower Everything To Perceive Intelligently
With a mission statement of “We Empower Everything To Perceive Intelligently”, STRADVISION is putting all of our effort to make better life for everyone through AI-based camera perception technology. Everyday, we focus on creating AI-based vision perception techonolgy with more than 300 members across 8 offices worldwide and we expect our software to perceive everything precisely & intelligently to make 1% difference in people’s lives. Thus, we are looking for members who would like to join our meaningful journey and face challenges that no one has done it before together at STRADVISION.
[Our Story]
• STRADVISION Seoul Office : https://buly.kr/Edu2Jdr
• STRADVISION’s Core Technology : https://buly.kr/74XS8bi
• STRADVISION's Welfare : https://buly.kr/BpFucdd
[Our Technology]
STRADVISION's Technology : https://stradvision.com/_ENG/dh/technology
• STRADVISION is the FIRST deep-learning based technology start-up company in the world who has obtained ASPICE CL2 certification in 2019.
• STRADVISION has also been honored with the AutoSens Awards for ‘Best-in-Class Software for Perception Systems’(Gold Award Winner) for 2 years in a row(2021, 2022).
• STRADVISION’s outstanding technology was recognized worldwide by successfully completing the Series C funding at KRW 107.6 billion with Aptiv and ZF Group in August, 2022.
• About 167 patents related to autonomous driving/ADAS have been acquired in Korea, Japan, US and Europe. As of today, STRADVISION is actively developing our technology to be differentiated.
• By the first half of 2025, vehicles equipped with SVNet surpassed 4 million units globally, maintaining growth despite economic slowdown and intensifying industry competition.
• In 2025, STRADVISION was ranked 10th in domestic AI competitiveness, following companies such as Samsung, Naver, and LG.
주요업무
[Mission of the Role]We’re hiring an Embedded AI Engineer to productize MV-Gen2 multi-camera, Transformer-based vision perception models (e.g., object detection / lane / occupancy), and E2E models(e.g., path planner, and control) across multiple automotive SoCs—Renesas, TI (Jacinto/TDA4), Qualcomm, and NVIDIA Orin/Thor.
You will lead INT8 quantization (PTQ/QAT), model porting, runtime/accelerator optimization, and—critically—provide HW-aware redesign guidance to internal DL model teams so architectures remain deployable and efficient under SoC constraints.
This role is a unique opportunity to work on high-impact, cutting-edge research that directly contributes to the development of next-generation autonomous driving systems.
[Key Responsibilities]
The selected candidate will be responsible for designing, developing, and optimizing deep learning models for ADAS/Autonomy.
1) Quantization & Deployment (Multi-Camera Transformers)
• Convert and deploy multi-camera Transformer perception models to embedded inference stacks using PTQ/QAT, calibration pipelines, and accuracy recovery strategies (mixed precision, selective quantization, layer-wise sensitivity).
• Build robust export/packaging flow (e.g., PyTorch ONNX target runtime) and resolve conversion/runtime issues (unsupported ops, precision constraints, graph transforms).
2) HW-Aware Model Redesign Feedback (Internal Technical Leadership)
• Provide actionable feedback to internal DL engineers on SoC-friendly Transformer design :
• Reduce attention/FFN compute & activation memory, manage token/BEV grid size, and avoid/replace ops that break target toolchains.
• Propose architecture patterns that preserve accuracy while meeting embedded constraints (bandwidth, SRAM/DDR pressure, operator support).
• Codify guidelines into “design rules” (allowed/avoid ops, preferred blocks, quantization-robust practices) and drive adoption through reviews and decision logs.
3) Multi-SoC Runtime & Accelerator Optimization
• Optimize latency/throughput/memory for real-time perception across NPU/GPU/DLA/DSP backends :
• Renesas: DRP-AI INT8 PTQ flow (calibration-based static quantization) and translation pipeline integration.
• NVIDIA Orin: TensorRT build/engine tuning, DLA constraints(INT8/FP16) and GPU–DLA workload partitioning.
• Qualcomm: ONNX Runtime QNN Execution Provider/ QNN SDK-based acceleration strategy and deployment validation.
• TI: TIDL quantization modes (PTQ/QAT) and “deployable-by-design” constraints.
4) Tooling & Performance Infrastructure
• Establish repeatable profiling + regression harness for latency/memory/accuracy across SoCs; publish performance dashboards and release artifacts.
• Collaborate with platform/firmware teams to unblock runtime integration and memory/IO bottlenecks.
자격요건
[Basic Qualifications]• MS + 3 years industry or PhD + 1 years (or equivalent), with 3–6+ years hands-on DL in total (vision/ML/robotics).
• Hands-on experience shipping DL models to embedded/edge devices (automotive/robotics/consumer edge), including INT8 quantization (PTQ/QAT) and performance tuning.
• Strong proficiency in Python + C/C++, model debugging, profiling, and deployment automation.
Practical knowledge of at least one deployment stack: TensorRT, ONNX Runtime, QNN, TIDL, DRP-AI, TVM/compilers, etc.
• Experience optimizing Transformer-based vision models (attention bottlenecks, activation memory, mixed precision, operator constraints).
