AI Engineer @ DeltaX — building perception systems for autonomous driving and in-cabin monitoring.
Computer Vision • Object Detection • Driver Monitoring Systems (DMS)
Edge AI — model quantization, ONNX conversion, and deployment to embedded devices.
Generative Models & Diffusion researcher (M.S., Soongsil University 🇰🇷)

I am a Machine Learning researcher and AI Engineer working at the intersection of computer vision, generative modeling, and edge deployment. My research focuses on Diffusion Models and GANs for high-quality image generation, and I completed my Master’s in Computer Science at Soongsil University, Seoul, under the guidance of Prof. Jaeyong Choi. Since December 2025, I have been an AI Engineer at DeltaX (Korea), where I build perception systems for autonomous driving and in-cabin monitoring — taking models all the way from training in PyTorch to quantized artifacts running on embedded automotive hardware. I am passionate about generative AI, computer vision, and bringing models from research into real-world deployment.

💼 Current Role

AI Engineer — DeltaX (Dec 2025 – Present)

• Automotive AI: Engineering end-to-end computer vision pipelines and post-processing logic for next-generation smart vehicle In-Cabin Monitoring Systems (ICMS).
• In-Cabin Monitoring System (IMS/DMS): Developing high-accuracy AI features to detect seatbelt compliance, track driver distraction (looking away, phone use, or hands off the wheel), and ensure passenger safety (identifying adults, children, and pets).
• Edge Deployment: Converting deep learning models to run efficiently on embedded devices, including model quantization (INT8 / 16-bit) and optimization for real-time, low-power inference.
• Hardware/Software Bridge: Working across the ML + embedded system — from training in PyTorch to compiled artifacts on target automotive hardware.

🏆 Selected Achievements

• 📄 4 peer-reviewed publications — The Visual Computer (Springer), KSII TIIS, IEEE ICOIN 2025 & MDPI Electronics
• 🚗 Production automotive AI — in-cabin monitoring (DMS) pipelines deployed to embedded SoCs at DeltaX
• 🎓 M.S. GPA: 4.16 / 4.5 at Soongsil University, Seoul
• 🥇 Top graduating student — B.E. Software Engineering, Mehran University (2023), GPA: 3.73 / 4.0
• 📊 SOTA results on DK-Font: SSIM 0.857, FID 10.45, surpassing Diff-Font and MX-Font
• 🌏 Multilingual font generation across Korean, Chinese, and Latin scripts from 3–5 reference glyphs

🎓 Education

• M.S. in Computer Science, Soongsil University, Seoul, South Korea (2023--2025)
  GPA: 4.16/4.5
  Researcher at System Software Lab under Prof. Jaeyong Choi
  Thesis: Diffusion-Driven Image Generation with Disentangled Style and Structure-Aware Fidelity
• B.E. in Software Engineering, Mehran University of Engineering and Technology, Pakistan (2018--2023)
  GPA: 3.73/4.0
  Graduated as the top student in the class
  Final Year Project: Real-Time Face Recognition Attendance System Using Computer Vision

🔬 Research Interests

• Image Generation (Diffusion Models, GANs)
• Multilingual Font Generation
• Few-shot and Zero-shot Learning
• Style Encoding and Transfer
• Computer Vision & Deep Learning
• Object Detection & Semantic Segmentation
• Human-Centered Design Tools using AI and Computer Vision

🧩 Core Competencies & Domain Expertise

🚗 Intelligent Vehicle Perception & Edge AI

• Advanced In-Cabin Systems (ICMS): Architecting real-time vision pipelines for occupant detection, 3D body keypoints, gaze tracking, and safety compliance (seatbelt, HOD, phone-use, CPD).
• Hardware-Aware Optimization: Porting deep learning models to resource-constrained embedded automotive hardware via INT8 / FP16 post-training quantization and custom calibration.
• Target Runtime Compilation: End-to-end framework translation across the PyTorch → ONNX → TI TIDL pipeline for low-power, edge-accelerated inference.

🎨 Generative AI & Image-to-Image Synthesis

• Image-to-Image (I2I) Translation: Designing deep learning architectures to translate, map, and transform visual content across completely different domains while preserving core structural integrity.
• Diffusion & GAN Architectures: Implementing state-of-the-art conditional diffusion models and Generative Adversarial Networks (GANs) for high-fidelity image generation, conditional synthesis, and advanced image editing.
• Synthetic Data Engineering: Developing automated pipelines utilizing foundation models (SAM) for high-fidelity image inpainting, synthetic data augmentation, and automated dataset annotation.

⚙️ Edge AI & Model Optimization

I focus on taking models from research to real hardware:

• Model Conversion: PyTorch → ONNX → hardware-optimized artifacts.
• Quantization: Post-training quantization (INT8 / 16-bit) with calibration for minimal accuracy loss.
• Embedded Deployment: Compiling & benchmarking models for TI TDA4VM and similar SoCs (AM68A, AM69A, AM62A).
• Optimization: Inference-time tuning, memory-bandwidth reduction, and real-time latency / FPS benchmarking.

📚 Publications

• Sami-Font: Structure and Style-Aware Multi-Scale Infusion for One-Shot Multilingual Typeface Generation
  First Author — The Visual Computer (Springer)
  Paper Link
• DML-Font: Multilingual Font Generation Based on Diffusion Model
  First Author — KSII Transactions on Internet and Information Systems (TIIS)
  Paper Link
• Positional Component-Guided Hangul Font Image Generation via Deep Semantic Segmentation and Adversarial Style Transfer
  Third Author — Electronics (MDPI)
  Paper Link
• Text-Conditioned Diffusion Model for High-Fidelity Korean Font Generation
  First Author — IEEE (ICOIN 2025)
  Paper Link

🛠️ Projects

DK-Font: Diffusion-Driven Multilingual Font Generation with Phonetic Awareness and Iterative Refinement
Abdul Sami, Jaeyong Choi
Tools: PyTorch, Diffusion Models, U‑Net, VGG‑19, ResNet, CLIP

• Developed a diffusion model for font synthesis across Korean, Chinese, and Latin scripts.
• Used phonetic-aware encoding and iterative refinement to enhance structural accuracy and style consistency.
• Achieved SOTA results in SSIM, FID, and LPIPS, surpassing prior work like Diff‑Font and MX‑Font.
• Implemented few-shot capabilities to synthesize full font sets from just 3–5 reference glyphs.
• Code & demos: DK-Font repository on GitHub

Unified Diffusion Model with Multi-Scale Style Infusion and Structure-Aware Losses (Ongoing)
Tools: PyTorch, Diffusion Models, Sobel Filtering, CLIP, VGG, U-Net

• Developing an enhanced diffusion-based font generation framework with unified single-phase training.
• Introduced Multi-Scale Style Infusion to inject style representations at encoder, bottleneck, and decoder stages.
• Integrated Sobel-based structural consistency loss to enforce stroke-level preservation during generation.
• Employed CLIP-based style loss for perceptual alignment between reference and generated glyphs.
• Designed for high-fidelity, structure-aware font synthesis across multilingual scripts.

Real‑Time Face Recognition Attendance System
Tools: Python, OpenCV, Face Recognition, SQLite, Tkinter

• Built a desktop UI that recognizes faces from video and automatically logs attendance.
• Features include face registration, verification, live video feed, and database integration.

Hangul Font Classifier
Tools: PyTorch, AlexNet, torchvision

• Implemented a CNN to classify 2,780 Hangul characters across font styles.
• Achieved ~95% accuracy in real-time image-based prediction.

💻 Technical Skills

• Deep Learning Frameworks: PyTorch, TensorFlow, Keras
• Languages: Python (fluent), C++ (basic), HTML/CSS (for fun)
• ML Tools: HuggingFace Diffusers, VGG Feature Extractors, OpenCV, NumPy
• Model Types: Diffusion Models, GANs, Style Encoders, CNNs, Object Detectors
• Edge AI & Deployment: ONNX, TI TIDL, INT8 / FP16 Quantization, TDA4VM & AM6xA SoCs, Embedded Inference
• Data Tools: Pandas, Matplotlib, Jupyter, Weights & Biases
• Other Tools: Git, Docker, Linux, LaTeX

🧑‍🏫 Teaching & Talks

• Guest Speaker – "Modern Diffusion Models for Image Generation", System Software Lab Seminar, Soongsil University (2024)
• Teaching Assistant – "Deep Learning Programming", Soongsil University (Fall 2023)
• Presenter – "Font Generation using AI", Internal Lab Meeting (2024)

🔗 Get in Touch

• 📧 Email: abdulsamimahar001@gmail.com
• 🌐 Portfolio: abdulsami101.github.io
• 📍 Based in Seoul, South Korea
• 💼 Open to PhD opportunities (Fall 2026 / Fall 2027)

💡 I'm open to collaborations in generative AI and creative deep learning applications. Whether you're working on a new idea, looking for a partner in research, or just curious about fonts and image generation — let’s connect!

🎯 Actively seeking PhD positions starting Fall 2026 / Fall 2027 in Computer Vision, Generative Models, and Multilingual AI. Open to research collaborations and industry roles in generative AI.