About Sergen
Need a Computer Vision or Deep Learning specialist who doesn't just train models but actually ships them?
I build production-grade perception systems end-to-end; from data pipeline and model architecture through to optimized inference on embedded hardware. With nearly 4 years in automotive AI, I bring the rare combination of ML research depth and real deployment know-how.
What I typically take on:
— Design and optimization of real-time CV pipelines (6DoF pose estimation, segmentation, object/keypoint detection)
— Multi-task architectures and domain-adaptive training frameworks (synthetic-to-real)
— Hardware-aware model optimization for edge inference (ONNX export, runtime validation)
— Camera calibration, 3D–2D projection consistency, and geometry-aware training
— Prototype to production: from training script to deployed, validated system
What sets me apart: I have a peer-reviewed publication in out-of-distribution detection and work daily at the intersection of research and production — where many ML engineers stop, I'm just getting started.
Python · PyTorch · OpenCV · C++ · ONNX · Docker · MLflow · Git · Linux
German
Native or bilingual
English
Fluent
Turkish
Fluent
Spanish
Conversational
Remote only
Primarily works remotely
Experience
- AptivComputer Vision & AI Research EngineerAUTOMOBILEJuly 2022 - Today (3 years and 11 months)Wuppertal, NW, Germany-Own end-to-end development of a deep learning–based Driver Monitoring System (DMS) model, including alignment of Python training and C++ inference pipelines to ensure production consistency-Drove significant KPI improvements across the DMS model, enabling production-level maturity and improving downstream features such as drowsiness detection-Replaced a classical pose estimation algorithm with a custom neural network architecture using PyTorch, achieving ~30% inference runtime reduction on target embedded hardware while maintaining accuracy-Built domain adaptation training framework bridging synthetic and real data, improving target-domain KPI by ~40%-Engineered multi-head neural network unifying sequential tasks, replacing a two-model pipeline and rule-based post-processing with a single forward pass and reducing end-to-end latency by ~15%-Implemented coordinate frame transformations and intrinsic parameter adjustment to ensure 3D–2D projection consistency under image undistortion, augmentation, cropping, and rescaling (OpenCV)-Conducted semantic segmentation research mapping monocular camera inputs to structured spatial representations-Generated semantic segmentation training labels from LiDAR point clouds via voxelization and DBSCAN clustering-Architected a multi-task training pipeline with masked loss formulation and GradNorm-based weighting across disjoint label spaces, enabling stable joint optimization and shared backbone representations
- Akka TechnologiesSoftware Engineering ConsultantAUTOMOBILENovember 2019 - December 2020 (1 year and 1 month)Munich, Germany- Developed analytical and visualization tools in MATLAB for automotive measurement data- Refactored legacy tools using object-oriented design principles to improve maintainability and modularity- Modified a C++ binary parsing module to extract and transform structured byte-level measurement data
Recommendations
Be the first to recommend Sergen
Help this freelancer shine by sharing your experience working together.
These freelancer profiles also match your criteria
Agatha Frydrych
Backend Java Software Engineer
4.7
(3)
2
Baptiste Duhen
Fullstack developer
4.6
(4)
5
Amed Hamou
Senior Lead Developer
4
(2)
7
Audrey Champion
Web developer
4.3
(3)
4
Education
- Master of ScienceRWTH Aachen University2021Electrical Engineering, Information Technology and Computer Engineering
- Study Abroad SemesterQUT (Queensland University of Technology)2019Management, Data Analysis, Managerial Finance