The third quarter of 2025 - Redesigning My Life Through LLMs
*A Challenging 2nd and 3rd Quarter of 2025
Unlike the first quarter, which was more satisfying than expected, the second and third quarters left me with some regrets.
In particular, even though I managed to quickly build a system using Ollama–LangChain–OpenWebUI, I hadn’t anticipated that unforeseen political issues within related departments would hold things back. It’s unfortunate that I can’t go into detail about this on the blog, but I’ll set those thoughts aside and continue with the main topic of this post.
Diving into LLM Service Development
Starting in March, I focused on deepening my understanding of LangChain and using it for service development. Embracing new technology is never easy. For example, I recently tried Cursor AI, which has been drawing a lot of attention, and found it less straightforward than expected. Without very detailed prompts, the generated code didn’t match my intentions. Even when I wrote detailed prompts, I often spent more time reviewing and improving the generated code. Perhaps I’m still not proficient with Cursor AI, but I believe that with time, it will mature into a highly capable tool.
As I became absorbed in service development, I found myself neglecting my blog writing. I tried not to let it fade into a short-lived attempt, but in hindsight, it feels like my main weakness was time management. 😫
GitHub Contributions Graph
LangChain’s relatively gentle learning curve made it easy to pick up quickly. However, I’ve come to realize that building a structure suitable for real-world tasks—and maintaining it—was far more challenging than I expected. Fortunately, I discovered that LangGraph could compensate for many of these limitations. Learning LangGraph alongside system development has taught me a lot. Although I still have much to improve, I’ve come to a clear realization that this is an essential step toward implementing AI Agents.
Meanwhile, I developed my own service using Ollama, OpenRouter (if I had known about it earlier, I might not have chosen Ollama…), LangChain/LangGraph, and OpenWebUI. While there’s still much room for improvement, I believe that time will resolve many of these shortcomings. At the very least, I feel that I’ve laid the foundation for my 2026 projects.
In Closing
The second and third quarters were a time when I was able to elevate my LLM skills to the next level. Although there’s still much to learn, I look forward to seeing a different version of myself in the fourth quarter when I start implementing AI Agents and working on MCP/A2A.
I hope to carry this same passion and focus into the next quarter. And to you, reading this post—I sincerely wish that you, too, continue to run boldly along your own path.