1// Auto-generated THOUGHTS_DATA 2const THOUGHTS_DATA = [ 3 { 4 "id": "thought-1782502180954", 5 "text": "Why cant we do the pre processing step of LLMs in local machines, such as extraction of texts from documents, so that we dont rely on their servers and waste our file upload limits? Had this idea for a standalone desktop app. It uses our local CPU to do all the extraction and data cleaning, then gives out a highly detailed markdown prompt. We just copy and paste that into our favourite LLM. We save limits and let the cloud GPUs focus entirely on the actual work!!", 6 "color": "#fdcfe8", 7 "date": "June 27, 2026" 8 }, 9 { 10 "id": "introductory-note", 11 "text": "Welcome to my Thoughts canvas! This is a dedicated, informal space for tiny thoughts, project ideas, and sometimes act as a tiny diary as well :) Each card is rendered as a colorful sticky note, managed locally by my custom offline CMS.", 12 "color": "#cbf0f8", 13 "date": "May 31, 2026" 14 } 15];
Line numbers count LF bytes from the start of the resource, as the search results do. Vendor segments are library code the classifier recognised; they are stored but not indexed. Bytes are shown as Latin1 characters, one per byte.