A personal recap of 2025: on running, LLMs, family, coffee, work (dimitarmisev.com)

In a reflective post about 2025, an individual shares their journey through personal and professional growth, highlighting their experiences with running, family, and advancements in AI. After grappling with knee injuries for years, they established a sustainable running routine that blends easy and interval training. This method has led to improved performance, with personal bests in 5k and 10k races, alongside a renewed appreciation for the outdoors. The author contrasts their physical endeavors with their burgeoning interest in large language models (LLMs), which shifted from a distant curiosity to a practical focus after the release of more advanced models like DeepSeek's V3. Significantly, the author delves into the technical challenges and triumphs of setting up a powerful NVIDIA GPU to train and run various LLMs. They successfully navigate issues related to hardware compatibility and cooling systems, which deepens their understanding of AI tools and frameworks. Furthermore, they engage in local experimentation with models, emphasizing the balance between private use and the cost-effectiveness of running AI systems at home. Their diversification into AI parallels their family life as they teach their six-year-old foundational skills, creatively fostering curiosity and critical thinking, all while building a cluster of Raspberry Pi devices for advanced database experiments, showcasing a fusion of personal passion and professional inquiry in the AI/ML landscape.