<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>System2: new entries</title><description>New projects and code built on TypeSafe&apos;s System One models. An unofficial directory.</description><link>https://systemtwo.dev/</link><item><title>pg-jev</title><link>https://systemtwo.dev/e/realzachi-pg-jev/</link><guid isPermaLink="true">https://systemtwo.dev/e/realzachi-pg-jev/</guid><description>A PostgreSQL extension that lets you write a WHERE clause in plain language: jev(people, the name is European) is an ordinary boolean function, so it joins, groups and sorts like any other. Each row is judged by Jev, batched many rows per request and cached per row, with jev_prob, jev_choice and jev_score for ranking and classifying. No embeddings, no vector column.</description><pubDate>Fri, 18 Sep 2026 06:09:12 GMT</pubDate><category>noul</category><category>choice</category><category>score</category><category>search</category><category>library</category></item><item><title>Jev Detector</title><link>https://systemtwo.dev/e/jev-detector/</link><guid isPermaLink="true">https://systemtwo.dev/e/jev-detector/</guid><description>A free AI-slop detector. Paste up to 10,000 words and Jev checks every sentence against known AI-writing patterns, then highlights the ones worth a second look with the pattern it matched, a strength score and a confidence. Framed as a writing aid rather than a verdict, with the Wikipedia rubric it draws on linked. No login.</description><pubDate>Fri, 18 Sep 2026 05:49:39 GMT</pubDate><category>classification</category><category>score</category><category>app</category></item><item><title>Jev Moderation Bot</title><link>https://systemtwo.dev/e/brainstormity-jev-moderation-bot/</link><guid isPermaLink="true">https://systemtwo.dev/e/brainstormity-jev-moderation-bot/</guid><description>A Discord moderation bot in Python that asks Jev whether each message is spam or a scam link, deletes offenders on the spot and escalates from a warning to a timeout. A /profile command scores a member across scam risk, spamminess, toxicity and helpfulness, and pardons are kept as precedents so the same false alarm is not repeated.</description><pubDate>Fri, 18 Sep 2026 05:44:11 GMT</pubDate><category>moderation</category><category>guardrails</category><category>classification</category><category>score</category><category>app</category></item><item><title>skillranker</title><link>https://systemtwo.dev/e/dicklesworthstone-skillranker/</link><guid isPermaLink="true">https://systemtwo.dev/e/dicklesworthstone-skillranker/</guid><description>A standalone Rust CLI that puts TypeSafe.ai&apos;s Jev at the center of skill selection: Jev evaluates your agent&apos;s live context, compares the available skills, and estimates which ones fit the next step. SkillRanker supplies the session integration, local safeguards, and inspectable feedback around it.</description><pubDate>Fri, 18 Sep 2026 02:25:47 GMT</pubDate><category>classification</category><category>guardrails</category></item><item><title>Jev Ultrafast</title><link>https://systemtwo.dev/e/browser-use-jev-ultrafast/</link><guid isPermaLink="true">https://systemtwo.dev/e/browser-use-jev-ultrafast/</guid><description>A browser agent from Browser Use where Jev decides every step. From a numbered table of the page&apos;s elements it picks the operation and the target in a single request, and a small LLM is called only when text has to be typed. Its demo runs a Google Flights search from one plain-English goal in about seven seconds.</description><pubDate>Thu, 17 Sep 2026 07:04:56 GMT</pubDate><category>choice</category><category>agents</category></item><item><title>Autoresearch feature discovery</title><link>https://systemtwo.dev/e/autoresearch-feature-discovery/</link><guid isPermaLink="true">https://systemtwo.dev/e/autoresearch-feature-discovery/</guid><description>A loop that proposes new questions, turns free text into numeric features with them, and uses a regressor&apos;s mistakes to decide what to ask next.</description><pubDate>Thu, 17 Sep 2026 00:00:00 GMT</pubDate><category>score</category><category>noul</category><category>machine-learning</category></item><item><title>Classification using confidence</title><link>https://systemtwo.dev/e/classification-using-confidence/</link><guid isPermaLink="true">https://systemtwo.dev/e/classification-using-confidence/</guid><description>Sort company annual reports into fine-grained industry groups with one Choice each, and use the answer&apos;s own confidence to fall back to the broader division when the model is unsure.</description><pubDate>Thu, 17 Sep 2026 00:00:00 GMT</pubDate><category>choice</category><category>classification</category><category>confidence</category></item><item><title>Classifying RAG passages</title><link>https://systemtwo.dev/e/classifying-rag-passages/</link><guid isPermaLink="true">https://systemtwo.dev/e/classifying-rag-passages/</guid><description>Judge each retrieved passage before it reaches the answering model: keep what helps, flag what contradicts the question, and drop anything carrying a hidden instruction.</description><pubDate>Thu, 17 Sep 2026 00:00:00 GMT</pubDate><category>noul</category><category>rag</category><category>guardrails</category></item><item><title>Date extraction</title><link>https://systemtwo.dev/e/date-extraction/</link><guid isPermaLink="true">https://systemtwo.dev/e/date-extraction/</guid><description>Pull absolute and relative dates out of documents by asking for the parts that are named, then resolving and validating them in code, with low-confidence cases sent for review.</description><pubDate>Thu, 17 Sep 2026 00:00:00 GMT</pubDate><category>choice</category><category>extraction</category><category>confidence</category></item><item><title>Double-checking citations</title><link>https://systemtwo.dev/e/double-checking-citations/</link><guid isPermaLink="true">https://systemtwo.dev/e/double-checking-citations/</guid><description>Check whether a quoted source really backs the claim it is attached to. One question gives the verdict, and its confidence decides which citations a person should look at.</description><pubDate>Thu, 17 Sep 2026 00:00:00 GMT</pubDate><category>choice</category><category>verification</category><category>rag</category></item><item><title>Function calling</title><link>https://systemtwo.dev/e/function-calling/</link><guid isPermaLink="true">https://systemtwo.dev/e/function-calling/</guid><description>Turn a plain-English request into a call to an ordinary typed function. Each function name and each closed-set argument becomes a question, so the answer is always something your code can dispatch.</description><pubDate>Thu, 17 Sep 2026 00:00:00 GMT</pubDate><category>choice</category><category>agents</category><category>routing</category></item><item><title>Guardrails for LLMs</title><link>https://systemtwo.dev/e/guardrails-for-llms/</link><guid isPermaLink="true">https://systemtwo.dev/e/guardrails-for-llms/</guid><description>Screen what goes into and comes out of an LLM app in a single request: describe the hazards, score how bad complying would be, and let your code decide to pass, review, block or reroute.</description><pubDate>Thu, 17 Sep 2026 00:00:00 GMT</pubDate><category>noul</category><category>score</category><category>guardrails</category></item><item><title>Hierarchical classification</title><link>https://systemtwo.dev/e/hierarchical-classification/</link><guid isPermaLink="true">https://systemtwo.dev/e/hierarchical-classification/</guid><description>Classify documents into deep category trees, such as patents, retail products, biomedical topics and source code, by running a beam search over the probabilities of each level.</description><pubDate>Thu, 17 Sep 2026 00:00:00 GMT</pubDate><category>choice</category><category>classification</category></item><item><title>Knowledge graph entity alignment</title><link>https://systemtwo.dev/e/knowledge-graph-entity-alignment/</link><guid isPermaLink="true">https://systemtwo.dev/e/knowledge-graph-entity-alignment/</guid><description>Decide which candidate pairs from two product catalogues are the same thing. A single Score question with three levels maps straight onto the three actions: merge, leave apart, or send to a curator.</description><pubDate>Thu, 17 Sep 2026 00:00:00 GMT</pubDate><category>score</category><category>noul</category><category>matching</category></item><item><title>Line-by-line search</title><link>https://systemtwo.dev/e/line-by-line-search/</link><guid isPermaLink="true">https://systemtwo.dev/e/line-by-line-search/</guid><description>Semantic search over a long terms-of-service document in one request: a Choice picks the line that answers a plain-language query, and a Noul checks whether the document answers it at all.</description><pubDate>Thu, 17 Sep 2026 00:00:00 GMT</pubDate><category>choice</category><category>noul</category><category>search</category></item><item><title>Parallel questions</title><link>https://systemtwo.dev/e/parallel-questions/</link><guid isPermaLink="true">https://systemtwo.dev/e/parallel-questions/</guid><description>Ask a whole briefing&apos;s worth of questions about one long document in a single call instead of one call each. Same answers, at a fraction of the cost and time.</description><pubDate>Thu, 17 Sep 2026 00:00:00 GMT</pubDate><category>noul</category><category>choice</category><category>score</category><category>cost</category></item><item><title>Pre-parsed value extraction</title><link>https://systemtwo.dev/e/pre-parsed-value-extraction/</link><guid isPermaLink="true">https://systemtwo.dev/e/pre-parsed-value-extraction/</guid><description>Let a regex find every candidate email, phone number or amount, then have the model select the right one. The output is always a verbatim value from the text: select, do not generate.</description><pubDate>Thu, 17 Sep 2026 00:00:00 GMT</pubDate><category>choice</category><category>noul</category><category>extraction</category></item><item><title>Re-ranking</title><link>https://systemtwo.dev/e/re-ranking/</link><guid isPermaLink="true">https://systemtwo.dev/e/re-ranking/</guid><description>Take a keyword-search shortlist and ask one question per query and candidate pair to reorder it. Worked through on legal case retrieval, where it lifts the right passage towards the top.</description><pubDate>Thu, 17 Sep 2026 00:00:00 GMT</pubDate><category>noul</category><category>search</category></item><item><title>SDE cascade</title><link>https://systemtwo.dev/e/sde-cascade/</link><guid isPermaLink="true">https://systemtwo.dev/e/sde-cascade/</guid><description>A two-stage cascade for structured data extraction: a small model extracts, a verification step checks, and only the doubtful cases go to a large reasoning model. Most of the quality for a fraction of the cost.</description><pubDate>Thu, 17 Sep 2026 00:00:00 GMT</pubDate><category>noul</category><category>extraction</category><category>cost</category></item><item><title>Self-consistency: choices</title><link>https://systemtwo.dev/e/self-consistency-choices/</link><guid isPermaLink="true">https://systemtwo.dev/e/self-consistency-choices/</guid><description>Add an explicit uncertain outcome to moderation decisions, then weigh how often labels agree against how much gets actioned automatically.</description><pubDate>Thu, 17 Sep 2026 00:00:00 GMT</pubDate><category>choice</category><category>confidence</category><category>moderation</category></item><item><title>Self-consistency: nouls</title><link>https://systemtwo.dev/e/self-consistency-nouls/</link><guid isPermaLink="true">https://systemtwo.dev/e/self-consistency-nouls/</guid><description>Send borderline probabilities to a human reviewer instead of forcing a yes or no, while keeping the underlying values in view so the routing stays explainable.</description><pubDate>Thu, 17 Sep 2026 00:00:00 GMT</pubDate><category>noul</category><category>confidence</category><category>moderation</category></item><item><title>Skill suggestion</title><link>https://systemtwo.dev/e/skill-suggestion/</link><guid isPermaLink="true">https://systemtwo.dev/e/skill-suggestion/</guid><description>Pick at most one skill for an agent&apos;s turn from a catalogue of well over a hundred. One request ranks them all and asks whether any is needed; a second reads the top few closely and may reject them all.</description><pubDate>Thu, 17 Sep 2026 00:00:00 GMT</pubDate><category>choice</category><category>noul</category><category>agents</category></item><item><title>Structure recovery</title><link>https://systemtwo.dev/e/structure-recovery/</link><guid isPermaLink="true">https://systemtwo.dev/e/structure-recovery/</guid><description>Rebuild Markdown from plain text that lost its formatting. One request rejoins hard-wrapped lines; a second labels each block as heading, list, code or callout.</description><pubDate>Thu, 17 Sep 2026 00:00:00 GMT</pubDate><category>choice</category><category>noul</category><category>formatting</category></item></channel></rss>