Research Engineer — AI, Deep Learning & Systems

📍 Incheon, South Korea · ✉️ [email protected] · 🔗 LinkedIn

AI Engineer with 5+ years of experience in deep learning, parallel computing, and system-level optimization across GPU-accelerated environments. I build and optimize AI models for real-time industrial applications; spanning multi-modal signal classification, video understanding, and large language model inference, with deep expertise in C++, Python, multi-GPU distributed training, and performance profiling for compute-intensive workloads.

🧪 Research

Positional-Encoding-Free Video Transformers — NoPE + GDN (paper under Review IEEE TCSVT)

Independent research on a video transformer architecture that removes explicit positional embeddings from video LLMs, pairing a No-Positional-Encoding (NoPE) design with a Gated DeltaNet (GDN) sequence backbone. The work targets stronger length generalization and higher throughput than rotary-embedding baselines for long-video understanding.

💼 Experience

Research Engineer · Sanion Co., Ltd.

Incheon, South Korea — March 2024 – Present

• Developed and optimized deep learning models for electrical grid diagnostics, achieving high-accuracy partial discharge classification using GPU-accelerated training pipelines
• Engineered a multi-modal fusion architecture combining a SAM (Segment Anything Model) ViT encoder with CNN branches to process 2D imagery and 1D electrical signals concurrently on multi-GPU setups
• Designed noise-removal and signal-denoising algorithms for partial discharge preprocessing, optimizing data throughput for model training
• Built a custom enterprise chat model using Qwen 7B with Retrieval-Augmented Generation (RAG), managing large-scale inference and context-window optimization
• Developed a substation automation system in C++, .NET Core, Redis, and gRPC for real-time IEC 61850 protocol communication and monitoring

Software Developer · Afiniti, Inc.

Lahore, Pakistan — December 2022 – March 2024

• Designed and launched new API methods for audio/video and text information, enhancing multi-modal product functionality across distributed microservices
• Optimized real-time audio/video streaming for 1,000+ concurrent agents in a multi-tenant system using WebSockets, reducing latency and improving throughput
• Integrated Whisper-based audio transcription into call-center products, enabling real-time speech-to-text for QA and analytics pipelines