01
Efficient AI Inference
I care about real deployment behavior, not only benchmark numbers. That includes latency, memory movement, and inference efficiency under practical constraints.
Research Portfolio / Systems-Oriented Builder
AI&SYSTEMS
AI Systems / Efficient Inference / Hardware-Aware Execution
AI Systems Researcher and Engineer
I study and build efficient AI systems across models, inference pipelines, and hardware-aware execution. My current interests include multimodal learning, CUDA optimization, runtime scheduling, and edge-oriented AI infrastructure.
Efficient InferenceCUDA & SystemsMultimodal AI
Beyond model usage, toward systems thinking.
I am interested in how AI models actually run in practice, including inference efficiency, memory behavior, runtime scheduling, and the coupling between models and hardware.
From CUDA kernel optimization and RAG-based code analysis to multimodal research and system-level inference thinking.
5+Selected technical projects
1Manuscript in preparation
50+Technical blog posts
3.7/4.0Master's CAP
Explore research and selected work01
Scroll
Yiming HuangResearch + EngineeringModel to Hardware
Focus
Research Focus
My current focus is how modern AI models can run more efficiently in real systems, especially under constraints of memory, latency, and hardware resources.
02
Compiler and Runtime Co-optimization
I am interested in treating quantization, operator fusion, code generation, and runtime scheduling as one connected systems problem.
03
Hardware-Aware Systems Thinking
My perspective is shaped by memory hierarchy, data locality, register pressure, instruction throughput, and the realities of edge and GPU platforms.
Work
Selected Projects
These projects show how I approach AI systems problems through both research reasoning and hands-on implementation.
01
AI SystemsCUDA GEMM Optimization and Architectural Analysis
Implemented and optimized GEMM kernels with shared memory tiling, register blocking, and profiling-guided analysis to improve arithmetic intensity and execution performance.
CUDAProfilingMemory Hierarchy
02
AI SystemsLLM + RAG Code Architecture Analysis System
Built a repository analysis system that combines LLMs, AST-based chunking, vector retrieval, and CUDA-aware parsing for structured code understanding.
LLMRAGAST
03
AI SystemsMultimodal Video Captioning Research
Designed a multimodal video-to-text pipeline with transformer-based alignment and efficiency-oriented system thinking.
MultimodalViTInference
Profile
Background
My background combines research-oriented study with practical engineering experience across data systems, maintenance, networking, and technical tool building.
Capability Thread
01-05- 01Master's research in multimodal AI and efficient inference
- 02Publication manuscript under preparation
- 03Engineering experience in data pipelines and automation
- 04Earlier systems and network operations experience
- 05Ongoing study in CSAPP, system programming, and performance analysis
Links
Materials
Currently focused on machine learning systems, low-level operator optimization, GPU programming, and AI infrastructure engineering.
