Hedi E. - AI Engineer

Ankara, Turkey

Experience

Dec 2024 - Present

1 year 1 month

Ankara, Turkey

Shipped a high-throughput video analytics platform handling 300+ surveillance streams (YOLO, MMDetection), ingesting 2TB+ daily with 24/7 reliability.
Served TensorRT-optimized models via NVIDIA Triton with dynamic batching; sub-100ms latency behind Dockerized REST services.
Instrumented Prometheus/Grafana dashboards (GPU utilization, throughput, tail latency) to reduce downtime by 40%.
Built SAM2+CVAT annotation workflow for rapid dataset bootstrapping from long-form video.
Led clinician-in-the-loop prototyping for CTA aneurysm detection; managed experiments on a self-hosted MLflow.

Jul 2022 - Sep 2022

3 months

Ankara, Turkey

Introduced DINO-based ViT matching loss for Pix2Pix, achieving 20% faster convergence with improved fidelity.

Apr 2022 - Jun 2024

2 years 3 months

Ankara, Turkey

Customized ResUNet++ for volumetric segmentation on 3D TOF-MRA; pipelines with PyTorch, MONAI, TorchIO.
Standardized MR processing with NiBabel, SimpleITK, MedPy (normalization, resampling, skull stripping).
Trained with AMP and cross-validation; achieved 85% sensitivity with 0.4 false positives per patient.
Experiment tracking and hyperparameter tuning via TensorBoard and Weights & Biases.

Mar 2022 - Jun 2024

2 years 4 months

Ankara, Turkey

ST360IQ: Spherical ViT-based no-reference IQA for 360°; PLCC up to 0.98, SRCC up to 0.97; ATSAL integration boosted metrics +2–3%, -20% runtime.
LGT360IQ: Local–global transformer with adaptive token selection; -17M params, +2% PLCC/SRCC across multiple benchmarks.

Sep 2021 - Dec 2021

4 months

San Francisco, United States

Built DETR-based localize+CNN classify pipeline in PyTorch for implant identification from X-rays; reduced manual workload by 43%.
Robust preprocessing using OpenCV/Albumentations; reporting with TensorBoard/Jupyter.