1. Submit the energy-saving analysis materials for December.
2. After debugging the voice recognition, extract the DeepSeek site.

【1】Think twice every day (think first about what you did well, think twice about what you did badly, think thrice about what you gained today and plan for tomorrow)：

Starting from January 20th, it took 4 days and 3 nights to finally implement the function of voice recognition, extracting from the DeepSeek site, and returning the results to the search box. Although it currently takes about 16 seconds to return, it has been achieved. During this process, I changed the voice recognition several times and tried different large models. After numerous debugging sessions, the function was finally realized. Without Dou Bao’s assistance in programming, it would have been impossible to achieve. Through repeated trials and errors, I found a compromise point for the hardware deficiency. The voice recognition was initially in the browser, then switched to Baidu Voice, and finally to Whisper loaded on the local server. The large model started with DeepSeek’s 7B model, which always timed out. The 1.3B model couldn’t be downloaded, so I finally switched to the qwen1.5B model under DeepSeek. It’s really exhausting. I keep debugging whenever I have time. However, the current memory consumption is a bit high and needs further optimization in the future.

【2】One sentence a day (one hundred words)：

Our lives are defined by opportunities – even the ones we miss.