<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>pin — the entire case file, answering</title><description>pin reads the entire criminal case file —scanned and handwritten— and hands back who is who, what happened when, and the page where it is recorded.</description><link>https://jjsutil.github.io/</link><item><title>The failure that should worry you about RAG isn&apos;t a missing answer. It&apos;s a fluent answer built from the wrong five paragraphs.</title><link>https://jjsutil.github.io/pin-landing/en/blog/how-rag-actually-works-and-where-it-silently-fails/</link><guid isPermaLink="true">https://jjsutil.github.io/pin-landing/en/blog/how-rag-actually-works-and-where-it-silently-fails/</guid><description>How retrieval-augmented generation actually works — chunking, embeddings, hybrid retrieval, re-ranking, context assembly — and the five distinct mechanisms by which it produces a plausible wrong answer instead of an honest &apos;not found.&apos; Grounded in the published literature, not in a vendor diagram.</description><pubDate>Thu, 03 Sep 2026 00:00:00 GMT</pubDate></item><item><title>When a model explains its reasoning, it isn&apos;t reporting what happened inside. It&apos;s generating more text.</title><link>https://jjsutil.github.io/pin-landing/en/blog/how-inference-and-interpretability-actually-work/</link><guid isPermaLink="true">https://jjsutil.github.io/pin-landing/en/blog/how-inference-and-interpretability-actually-work/</guid><description>How inference mechanically works in a modern LLM — the forward pass, the KV cache, sampling, quantization and what each one actually costs — and what interpretability research can and cannot currently tell you about why a model produced an output. Part 2 of our technical-depth series.</description><pubDate>Thu, 27 Aug 2026 00:00:00 GMT</pubDate></item><item><title>The failure that should worry you about OCR doesn&apos;t look like garbled text. It looks like 98% confidence and a missing paragraph.</title><link>https://jjsutil.github.io/pin-landing/en/blog/ocr-what-we-learned-and-what-the-literature-says/</link><guid isPermaLink="true">https://jjsutil.github.io/pin-landing/en/blog/ocr-what-we-learned-and-what-the-literature-says/</guid><description>We built our OCR pipeline as a confidence-routed cascade, benchmarked it against three newer document-understanding tools, then threw genuinely dirty scans at it and watched the safety net fail to catch four real bugs. What we found, and what the current literature actually supports.</description><pubDate>Thu, 20 Aug 2026 00:00:00 GMT</pubDate></item><item><title>I didn&apos;t start pin because I love legal technology. I&apos;d already watched this exact failure three times.</title><link>https://jjsutil.github.io/pin-landing/en/blog/why-i-started-pin/</link><guid isPermaLink="true">https://jjsutil.github.io/pin-landing/en/blog/why-i-started-pin/</guid><description>Grab, Uber, Stripe — three companies, three different products, the same wall: they scaled faster than their own paperwork could follow. Pin exists because the legal industry hit that wall and nobody built the fix.</description><pubDate>Thu, 13 Aug 2026 00:00:00 GMT</pubDate></item><item><title>No answer is better than the reading it came from.</title><link>https://jjsutil.github.io/pin-landing/en/blog/how-a-scanned-case-file-is-read/</link><guid isPermaLink="true">https://jjsutil.github.io/pin-landing/en/blog/how-a-scanned-case-file-is-read/</guid><description>Before a system can answer questions about your case file, someone has to turn the paper into text. That step is called OCR, it never shows up in a demo, and it decides the quality of everything that follows.</description><pubDate>Thu, 06 Aug 2026 00:00:00 GMT</pubDate></item><item><title>Your client trusted you with the case file. Not with your tool&apos;s servers.</title><link>https://jjsutil.github.io/pin-landing/en/blog/what-happens-to-your-case-file-when-you-upload-it/</link><guid isPermaLink="true">https://jjsutil.github.io/pin-landing/en/blog/what-happens-to-your-case-file-when-you-upload-it/</guid><description>In May 2025, a New York court ordered the preservation of even the chats users had deleted. The question is no longer what artificial intelligence does with your case file. It&apos;s where it ends up, who retains it, and under what rules.</description><pubDate>Thu, 30 Jul 2026 00:00:00 GMT</pubDate></item><item><title>A tool isn&apos;t measured by how often it&apos;s right. It&apos;s measured by what it costs to check.</title><link>https://jjsutil.github.io/pin-landing/en/blog/how-an-ai-answer-is-verified/</link><guid isPermaLink="true">https://jjsutil.github.io/pin-landing/en/blog/how-an-ai-answer-is-verified/</guid><description>The duty to verify is already written into the rules, and already being sanctioned. What no one explains is how you fulfill it without losing the time the tool promised to save you. That cost isn&apos;t set by your discipline: it&apos;s set by the architecture.</description><pubDate>Thu, 23 Jul 2026 00:00:00 GMT</pubDate></item><item><title>Nobody gets sanctioned for using artificial intelligence. They get sanctioned for signing what they didn&apos;t read.</title><link>https://jjsutil.github.io/pin-landing/en/blog/is-it-legal-to-use-ai-for-work/</link><guid isPermaLink="true">https://jjsutil.github.io/pin-landing/en/blog/is-it-legal-to-use-ai-for-work/</guid><description>A lawyer filed 38,477 briefs in one weekend, and civil court judges asked to ban AI from the courts. The question left for you isn&apos;t whether you can use it. It&apos;s what you&apos;re signing when you do.</description><pubDate>Thu, 16 Jul 2026 00:00:00 GMT</pubDate></item><item><title>It didn&apos;t forget your case file. It never read the whole thing.</title><link>https://jjsutil.github.io/pin-landing/en/blog/why-no-tool-has-worked-for-you/</link><guid isPermaLink="true">https://jjsutil.github.io/pin-landing/en/blog/why-no-tool-has-worked-for-you/</guid><description>You paid for the subscription, uploaded the case file, and the tool failed you in three different ways. None of the three was a glitch: all three are the same limit.</description><pubDate>Thu, 09 Jul 2026 00:00:00 GMT</pubDate></item><item><title>&quot;Artificial intelligence&quot; doesn&apos;t name a technology. It names a promise.</title><link>https://jjsutil.github.io/pin-landing/en/blog/what-machine-learning-actually-is/</link><guid isPermaLink="true">https://jjsutil.github.io/pin-landing/en/blog/what-machine-learning-actually-is/</guid><description>You&apos;re going to get sold &quot;artificial intelligence&quot; this year. Three words are enough to evaluate the pitch, and to know what you&apos;re signing when you sign.</description><pubDate>Thu, 02 Jul 2026 00:00:00 GMT</pubDate></item></channel></rss>