Eldor Ibragimov, PhD

ML Engineer | Computer Vision & Deep Learning Specialist

Seoul, KR.

About

Eldor Ibragimov, PhD, is a distinguished ML Engineer specializing in Vision-Language Models, multimodal learning, and deep learning, with a robust track record in translating cutting-edge research into scalable AI systems for real-world industrial applications. He excels at directing production rollouts, optimizing model inference for lightweight hardware, and delivering high-impact R&D projects that drive significant improvements in asset management and operational efficiency.

Work

SISTech.AI

|

Senior Computer Vision Research Engineer

Seoul, Seoul, Korea (Republic of)

→

Summary

As a Senior Computer Vision Research Engineer, Eldor directed the production rollout of the AI-based RoadVision platform across 13 provinces, integrating advanced computer vision pipelines to enhance road infrastructure assessment and reduce inspection time by 80%.

Highlights

Directed the production rollout of the AI-based RoadVision platform throughout 13 provinces, integrating PyTorch and Docker-based computer vision pipelines for comprehensive road infrastructure assessment.

Developed and deployed real-time object detection models for road surface analysis, achieving 95% mAP and reducing inspection time by 80%, informing data-driven asset management decisions.

Successfully delivered over 10 client-facing R&D projects in road infrastructure assessment through a combination of strong technical solutions and collaborative team leadership.

Optimized model inference for lightweight hardware (TensorRT, ONNX, PyTorch), enabling efficient deployment at scale for sustainability.

UDNS

|

Computer Vision Engineer

Seoul, Seoul, Korea (Republic of)

→

Summary

As a Computer Vision Engineer, Eldor engineered and deployed advanced segmentation and detection models for platforms like CrackViewer.com, achieving high accuracy in defect detection and significantly increasing real-time processing speed for highway monitoring.

Highlights

Adapted and fine-tuned UNet architecture for construction site damage segmentation, achieving 0.7 Dice loss and 0.9 accuracy, demonstrating transfer learning expertise.

Developed and deployed segmentation and detection models (mIOU of 0.8) for the CrackViewer.com platform, delivering a web-based solution integrated with AWS cloud services.

Designed a high-accuracy (99%) defect detection and classification algorithm for carbon material quality assessment, streamlining quality control processes.

Engineered and optimized a real-time road damage detection system using ONNX and TensorRT, achieving a 4× FPS increase (100 FPS) for large-scale highway monitoring.

Education

Sejong University

Seoul, Seoul, Korea (Republic of)
→
PhD and Master Structural Engineering Grade: 3.87/4.5

Tashkent Technical University

Tashkent, Tashkent, Uzbekistan
→
Bachelor Mining Grade: 3.82/4.0

Publications

Crack Segmentation With Text-guided Features Using Vision Language Model

REAAA

Research on crack segmentation utilizing text-guided features within a vision-language model framework for enhanced accuracy.

Automated Pavement Condition Index Assessment Using Deep Learning and Image Analysis: A Comprehensive Approach

Sensors

A comprehensive approach to automated pavement condition index assessment, leveraging deep learning and image analysis techniques.

A Vision-based System for Inspection of Expansion Joints in Concrete Pavement

Smart Structures & Systems

A study presenting a vision-based system designed for the precise inspection of expansion joints in concrete pavements.

Automated pavement distress detection using region based convolutional neural networks

International Journal of Pavement Engineering

A publication detailing the use of region-based convolutional neural networks for automated pavement distress detection, enhancing infrastructure assessment.

Certificates

Agentic AI

DeepLearning.AI

Google Project Management: Professional Certificate

Coursera

Deep Learning Nanodegree Program

Udacity

Data Analyst Nanodegree

Udacity

Deep Learning with Python and PyTorch

edX

Skills

Computer Vision

Deep Learning (PyTorch), Object Detection, Image Segmentation.

Generative AI

VLMs, LLMs, RAG, APIs (OpenAI, Claude), Agent Workflows, Multi-Agent Systems.

Software Development

Python, C++, Gradio, Django, OpenCV, Scikit-learn.

MLOps

Model Deployment (Docker), CI/CD Integration, Model Monitoring, Data Pipelines.

Leadership & Communication

Team Leadership, Project Management, Technical Communication, Problem-solving.

Projects

Vision Language Model (CLIP) for Crack Segmentation

→

Summary

Developed an innovative vision-language segmentation pipeline leveraging modified CLIP (VIT-L/14) and transformer architectures to enhance context-specific crack detection on diverse surfaces.

AI-Based Face Patch Defect Detection System for Smart Manufacturing

→

Summary

Engineered and deployed a highly accurate, real-time defect detection system for smart manufacturing using YOLOv8, significantly reducing defects in face patch production.

AI-Enhanced Road Surface Data Collection

→

Summary

Led the design and implementation of an efficient data labeling strategy and auto-labeling pipelines for road surface damage, ensuring high-quality and timely dataset completion.

RoadVision Integrated System for Road condition assessment

→

Summary

Designed and implemented YOLOv3-based object detection modules for the RoadVision system, enhancing automated road condition assessment with increased precision.

AI-Powered Crack Segmentation and Assessment Platform - CrackViewer.com

→

Summary

Developed a comprehensive AI-driven platform for concrete crack assessment, integrating Python, Django, and DeeplabV3+ to provide actionable insights for maintenance.

AI-Enhanced Pavement Assessment Platform Utilizing YOLO and PyTorch

→

Summary

Developed and deployed a web platform leveraging YOLOv5 and PyTorch for precise, automated pavement condition analysis, demonstrating full-stack development expertise.

Data Generation using GAN for Pavement Surfaces

→

Summary

Addressed pavement image dataset limitations by utilizing a GAN (TensorFlow) to significantly expand datasets and enhance model accuracy through data augmentation.

Real-Time Defect Detection on Carbon Material Production Line

→

Summary

Designed and engineered custom image processing algorithms for a carbon material production line, achieving high accuracy in defect detection and streamlining production processes.