Projects

I started exploring OpenClaw because I wanted hands-on experience with the kind of agentic systems that were drawing so much attention (even Jensen Huang talked about it in NVIDIA GTC 2026). Rather than just following the hype, I rented a Virtual Private Server (VPS) and set up my own system to understand what these tools can and cannot realistically do.

The goal was to build a personal agent workflow that could eventually support my day-to-day work: monitoring new research in areas relevant to me, helping manage parts of my website, assisting with scheduling and coordination, and later exploring whether similar workflows could support routine simulation and research tasks.

This project also became the bridge between my broader AI exploration and this website, since part of the long-term idea is to let the system help maintain and update parts of my online research presence.

I had wanted to build a personal website for a long time, but kept pushing it aside. With the recent rise of AI-assisted coding tools, I finally had a practical way to put it together without turning it into a massive side project (which is not my main area).

This site was built as a way to showcase my work, organize my research identity, and create a space that actually reflects my background across thermal engineering, simulation, software, and experiments. A big part of the process was learning how far modern AI coding tools can help with rapid development, iteration, and refinement when building something from scratch under realistic time constraints that I have with my own research.

The result is not just a website, but also part of a broader workflow experiment in how AI-assisted development can help individual researchers build useful systems quickly and independently.

A lot of my independent AI exploration has centered on one question: can these systems remove the repetitive labor around literature review and research organization without replacing the actual thinking? That question led me to build and test retrieval pipelines, workflow automation, and tool interfaces for research use cases.

The starting point was my own paper-writing workflow. I wanted a better way to search across research papers, recover specific ideas, group them around topics I was writing about, and reduce the manual effort of maintaining large reference spreadsheets and summaries. That led me to experiment with retrieval-augmented generation, research-aware pipelines, and tool-driven workflows connected to the way I already manage papers and notes.

I also explored automation around routine PhD and research tasks such as literature tracking, simulation-job handling, and document workflows using LLMs. Some of these explorations were eventually overtaken by more polished tools such as NotebookLM by Google, which I now use in practice, but building these systems myself was still valuable because it helped me understand the underlying workflow, the limitations, how LLMs work and what is actually useful for research.