AI Engineer · Agent Architecture
이용환
I build LLM-based agents for complex workflows, connecting domain knowledge, tools, and multi-agent orchestration. My work spans agent architecture, memory, retrieval, and context management.
Agent applications and research projects, with the architecture, implementation, and validation behind each.
AI Agents · 2026 · In progress
A conversational CAD agent that connects natural-language requests to modeling tools, geometry validation, and an interactive 3D view.
Natural language → CAD tools → verified geometry
AI Agents · 2025
Developing domain agents for workflow automation, with multi-agent orchestration, tool integration, and memory and context management.
Domain knowledge connected to agent execution
Reinforcement Learning · 2024
Teaching PostgreSQL join operators to learn during query execution and share information across operators.
Online learning inside the query engine
LLMs · 2026
A pure-PyTorch study implementation of the DeepSeek-R1 training recipes: R1-Zero, the multi-stage R1 pipeline, and distillation.
Group-relative advantages · no learned critic
LLMs · 2024
Adapting code language models with 4-bit quantization and LoRA, exploring dataset, context length, and loss choices.
13B model · 4-bit QLoRA adaptation
Reinforcement Learning · 2021
Implementing and comparing DQN with sequential and parallel experience collection, using a shared learner, replay memory, and separate evaluation.
Parallel collectors feeding a shared DQN learner
Statistical Learning · 2019
Estimating nine Poisson intensities with expectation maximization, with Fisher-information analysis and a reproducible, seeded simulation study.
40,000 seeded simulations · nine parameters
I am an AI Engineer developing domain agents for workflow automation in civil engineering and construction. I design how agents delegate tasks, use tools, validate results, and recover from failed runs. Previously, I studied Artificial Intelligence at Oregon State University, where I researched reinforcement learning and online learning for database query optimization. My projects also include conversational CAD and LLM fine-tuning.
Artificial Intelligence
Research in reinforcement learning and database query optimization
Implementations, experiments, and the ideas behind them.
Attention · Sequence Models · Machine Translation
LSTM · Sequence Modeling · POS Tagging
Word Embeddings · GloVe · PPMI
For conversations about agent systems, workflow automation, or applied machine learning.
kapshaul@gmail.com
