Model capabilities & reasoning
Exploring the limits of Transformers, context, memory, reasoning, and dynamic computation through experiments in real tasks.
Working around AI agents, knowledge systems, and software engineering. This is where I keep research notes, public projects, and ideas worth writing down.
CURRENT WORK
Conduct research and engineering practices centering on agents, real-world business logic, and software systems.
Exploring the limits of Transformers, context, memory, reasoning, and dynamic computation through experiments in real tasks.
Connecting model capabilities with planning, tools, permissions, memory, and validation so agents can reliably complete complex abstract tasks.
Connecting engineering documents, asset entities, digital delivery, and business rules to turn scattered material into traceable, queryable industrial knowledge.
Combining model serving, retrieval, workflows, data storage, evaluation, and observability into intelligent systems that can be deployed and maintained.
A PATH SO FAR
From algorithmic research to real-world industry applications, each step has been about turning understanding into systems that can be tested, used, and improved.
First encountered reinforcement learning and completed my undergraduate thesis with Q-learning.
Studied classical machine learning methods systematically and began academic research.
Contributed to a National Natural Science Foundation project, began working with LLMs, and entered engineering practice through an internship.
Completed academic and graduate theses on time-series models, worked on LLM fine-tuning in real projects, and began building AI products independently.
Working full-time on applied AI in traditional industries, refining systems that can make a difference in real settings.
PUBLIC PROJECTS
A close analysis of the concrete process behind the Shannon project.
RESEARCH · GITHUB ↗ 02A repository for useful research notes and references.
NOTES · GITHUB ↗ 03 · FORKAn open-source framework for document understanding, semantic retrieval, and RAG.
RAG · GITHUB ↗SELECTED NOTES