Triage & routing

家庭内AI秘書に話しかけるときに元々 haiku で意図分類させてから後段の sonnet/opus/fable に流してたのを Jev に置き換えて playground を作ってみた。タイピング進む度に意図分類が変わっておもろい https://t.co/WORQ03oZvA

I created a group for doctors appointsments, LinkedIn spam mails and my mail app now automatically puts my emails into this folders using jev. My inbox is just too cleaned up right now for a cool demo. guess i need to collect some more emails especially things like newsletters etc. Jev does not make any IMAP transactions, it's all just in the app Using my openrouter api key in the app. i totally f

JevとHaikuなど比較してGoogleニュース400本をタイトルから仕訳するシミュレーターをkote2ビルダーで作ってみた。最後に結果が出るよ。 https://t.co/Evv3b7Dnaa

I tried building a resume screener using @typesafeai 's Jev, not a plain LLM: An LLM writes a 1–10 score and a confident paragraph. Ask twice, get a new number. Jev answers one small question per requirement with a typed probability. Code ranks, unsure answers get flagged, every score is traceable. In short, consistent answer every single time. Check it out. Link in the comment.

話題のJevで、チャットでの発言をチームへの貢献の種類ごとに仕分けるデモを作りました！！ 「質問に答えた」「資料を共有した」など8種類を、スキャンしながらその場で判定してます。精度も高い。 1,000件を処理するのに約130秒・9円と早すぎ＆安すぎなのもすごい。。 https://t.co/Qt5I6vlNde

用 hugging face 上的 500 条真实电商客服数据做了一个电商客服分单工作台 让 JEV 和 DeepSeek 处理同样的 500 条测试工单：识别客户诉求，自动分到退款、物流、催发货等不同类别 结果是：JEV 用了约 83 秒，完成全部 500 条，花费0.01美金；DeepSeek 在这轮停单收尾后，完成了 173 条，花费了0.06美金 左边已经全部归档，右边才处理了三分之一左右，还贵了5倍 ！ 从 我们实际 FDE工程 的视角看 JEV 最牛逼的就是：以前一个具体业务里每天重复几万次的分类、判断、分流，会花费极大量的时间，但现在有了专门干这件事的模型 今天是客服分单，接下来可以是销售线索筛选、内容审核、业务路由……想象一下，把这些高频环节一个个接进系统，能释放多少效率？ 未来已致！AI 将真正完全的进入企业业务流程 提高10倍100倍效率！

最近 JEV 很火，我这边做了一些实测，做出了一个符合它模型能力的正确使用场景。 很多人不理解它为什么火，这里我来做一个解释：它就像是 GPU 渲染（就像你打游戏一样），而普通大模型就像 CPU，是一点一点输出的，玩过 Stable Diffusion 的人都应该明白这个概念。 回到正题，我写了一个 Demo，实际拿以前的工单让它去做判断。JEV 的输出非常快，可以迅速对这些工单进行整理与判断；而普通大模型需要一条一条输出并判断，响应速度会慢很多,这里我使用的是 GPT 5.6 Luna 和它做对比，在时间上差不多接近 60 倍的消耗 目前不适用的场景我也举个例子，比如Codex Computer Use 。因为你每一步思考都需要 OCR 去决定、去判断，然后再交给 GPT 这种大模型，JEV只是做单一的决策或者门禁判断，所以这种场景下它并没有优势，即使有提升也是很微弱的。它更适合的场景在

Wanted to something small and useful with viral Jev from @typesafeai So, i create Jev Issue Triage. Give it a public GitHub repo and it will pull all open issue, really fast. Also the token usage is really low. In this experiment we use AI as a decision engine, which is the use case of Jev. And more will come as we dicover it. Code is in my github repo - nabendu82

we gave jev 12000 synthetic records of Diabetes it was able to classify then in seconds jev is a classifier and it works insanely well when paired with an LLM which acts a rulebook creator https://t.co/qX9lGuOFBU

I wanted to see where Jev might fit into an accounting workflow, so I set up a test using 250 synthetic bank transactions. Each model got the same transaction details, a simplified chart of accounts, some bookkeeping rules and a written summary of the evidence. The job was to pick a category and decide whether the transaction needed an accountant to review it. So it was making decisions from infor

tested a homebrew Jev reflex gate locally on a 4GB GTX 1650 Ti: the premise: instead of burning slow thinking tokens on standard developer collisions, use a small 2B model as a sub-20ms System 1 decision gate. base 2B models hit an 83.3% catastrophic action rate (e.g. 57.0% probability of reformatting disk on a port 8080 collision). trained an 8.6MB LoRA adapter (4-bit NF4) into a JevMiniCPM refle

文章を書かないAIモデルが話題になってて 名前はJevっていうんだけど こっちが渡した選択肢から答えを選んで確率を返すだけなの 答えの形は3種類あって はいかいいえのどちらかと用意した候補のどれかと段階の点数なんだけど たとえば「この文は書き直すべきか」を聞くと はいの確率が0.91みたいな数字で返ってくるんだ 文章を作らないから速くて安くて 手元で同じ判定をClaudeと比べたら7.9秒が1秒かからなくなって 料金も出力にはかからないんだよね 実際にどこまで使えるか動画制作の流れの中で試してみたので 続きはYouTubeで https://t.co/IMQUFgOWNQ

jev お触り、最前面ウィンドウを細かく OCR で読んで、jev に遊んでるか仕事してるかを判定させてみる X 見ててもホーム画面はコンテンツがまあまあ仕事っぽいので遊び判定にはならず、「ゲーム」などで検索すると一気に遊び判定にできた（普通にゲームタイトルでググったりすると当然一気に振れる）。 しばらく触ってみて、日本語でも判定は安定しているが、初回の接続確立を除いても 1 判定 300ms 前後かかった。リージョンは選べなさそうなので、日本からだと今のところ仕方ないか。 jev のエンドポイントは一つしかなくて、その中で - noul（yes の確率を返す） - score（順序のある段階のどこに位置するかを返す） - choice（定義した選択肢から一つ選ぶ） の 3 つの type が使える。 確率のついた決まった構造しか返さないという割り切りなので、「高速・並列・安価・非ルールベ

Jevのユースケースとして会社HPのお問い合わせHPに導入して営業メールと判断したらお断りする機能を作ってみた📝 https://t.co/HPLkXuiTJ0

JevでWebサイトの表示を最適化するコンセプトを試してみた。リファラ・時刻・cookie・GETパラメータ（自由文）から「ユーザの目的は何か」を判定して、見出しとForm要素を差し替える。これによってユーザが求める情報を優先して表示できる。あらかじめ用意した部品から選ぶから品質担保できる点も。 https://t.co/uoqMC9kx5v

i got Jev inside a clinical workflow. this note mentions 6 conditions. OpenMed reads each span in context, Jev makes 6 typed decisions, and code lets 1 into the current problem list, blocks 4, and sends 1 to human review. https://t.co/ZeP59oZMpx

toBリードの業種判定にSonnetを使用していたけど、Jevに切り替えることでコストは200分の1、速度は1.5倍ほど早くなった https://t.co/x55m40he7k

I let JEV qualify 3,000 companies in milliseconds Scrape, enrich, ICP score, verify, CRM write, draft. JEV decided what runs next. 1,260 came out as fit. 61 it wasn't sure about, those went to a human for review. 41 seconds. $0.008. Agencies could be saving $150/m on heavy qualifications with their big target lists Live in Orbit for any GTM motion. Comment "JEV" and i'l send the link so you can tr

JEV is king at categorization! 🚨 JEV beats the TOP embedder (Qwen3-Embedding-4B) at categorization! 46% vs 97% accuracy! Not a simple win! We benchmarked +1000 TODO titles: - JEV scored 97% accuracy compared to Opus 5 reference. Qwen3 only 46%! - Embedder costed 0.15$/million TODO vs JEV 10$/million TODO Extremly cheap and extreme accuracy while staying superfast! Hopw you guys love it too! Blogpo

中途採用の書類選考について、JevとClaudeでそれぞれ評価し、処理速度の差を比較してみた 100人分の書類の判断に、Jevが4秒、Claude(Haiku)が20秒と5分の1の時間で完了 それだけでなく、コストはなんと1円、、、30分の1の値段でした それで評価ほぼ変わらずとなかなか利用余地ありそうです オカムラさんの動画を参考にさせていただきました！

Built Jev for X as a small Chrome extension. 🚀 It adds a live semantic layer directly on top of X: 🧠 Timeline posts get categorized 💬 Replies can be filtered in the context of the original post 🧩 Different conversations use different reply labels For example, an opinion thread might use: Agree · Disagree · Counterargument · Question While a product launch might use: Feature Request · Pricing Conce

用JEV做简历筛选这个案例确实有趣且实用，我自己一vibe 了这个实验。500份面试，每份纪要都让 Jev 分别回答 6 个问题，一共3000次判断，用时 31.4 秒，估算 $0.03173 一次请求详细字段见评论 https://t.co/mblXTmbska

Jev is a decision classification model. Here I'm using it to a/b test X post reactions on target demographics I swear this is not a fake jev demo. Try it out here: https://t.co/vFvkhEmGXg https://t.co/aQO7pqPmLm

I built an AI inbox that sorted 200 emails in 5 seconds. Jev handled the obvious ones and left me with just 23 to review. 500 emails reportedly cost just 3.5 cents. The AI does the boring decisions. You handle the uncertain ones. Comment “Agent OS” for the guide. https://t.co/2g7Ei2vMwy

1/4 TypeSafe AI has introduced a new "species" of models. The first is Jev: structured answers, not chat. I built HelloJev, a small PoC that sends data pipeline logs to Jev and asks what to do next: retry, fix, or investigate. Here is the demo. https://t.co/PUFQQN9XQG

been playing with @typesafeai Jev and honestly - insane. what a time to be a builder found a use case i couldn't stop building: instant model routing. why send every request to your biggest model? Jev scores the task in ~100ms, picks 1 of N, and returns a confidence + a needsReview flag. cheap calls stay cheap - only the uncertain ones escalate to claude/gpt. shipped it as a tiny 0-dep toolkit. je

jev'i elimdeki gerçek projenin içerisine soktum bakalım ne yapacak diye bi projemde 200k ürün var, 20 property üzerinden 41 soru sorup eleme yapıyoruz. ben şimdilik kategorilere göre ayrılmış bi 9k'lık csv ile oynuyorum videoda 200 ürünlük bir işlem yaptırıyorum. 12,7 saniye sürdü, 319 istek attı, 0.012 cent yazdı. 200k'nın hepsini döksem 13 dolar filan. şuana kadar jev benim açımdan kendini kanıt

Multi-second LLM classification is officially dead 💀 Just built a drag-and-drop agent router powered by @TypeSafeAI Jev. 1. Input comes in 2. Jev evaluates intent in ~15ms 3. Directs flow to the exact downstream agent Zero prompt parsing, What do you think? 👀 #BuildInPublic #AIAgents #TypeSafeAI #TypeScript

I used Jev (@typesafeai ) as the decision layer in a Pipecat voice pipeline. Three typed primitives, two calls per turn: Noul — a yes/no with a probability Choice — a labelled decision with your own criteria Score — a scalar on a rubric you define Call 1 runs before the LLM. Noul("is this a complete thought?") ends the turn semantically instead of on a silence timer, and Choice("which support flow

I tested Jev, @typesafeai's new classifier model, on 100 real banking support messages (77 intents, zero-shot): → 82% accuracy → 91.9% accuracy on the 74% of tickets where it was ≥90% confident → 329 ms median latency → $0.009 total Code: https://t.co/HtEjw82e7o https://t.co/zGGiGNGdJa

I made an experimental model router for @pidotdev using Jev as the decision maker. It's not a generic "pick a cheaper model" router. It isolates work into logical threads and prices cache reads/writes, cold context, expected output + the cost of switching back. @CompleteSkeptic @typesafeai This wouldn't of been possible without Jev, it's an amazing primitive to make software with 💪🏻

Using Jev from @typesafeai to improve matching performance, accuracy and cost in our automated @blackbaud RE NXT CRM integration. https://t.co/TVdfZOa3wI

Doc-OCR router using Jev @typesafeai A Jev-powered router that looks at a PDF page by page, decides which pages actually need OCR, extracts the rest locally. Result: save cost on # OCR pages + speed https://t.co/ZjXqHSjSGh

my video on the routing tool blew up on YT! i crossed 10k views in less than 19 hours, i still can't believe it - my most viewed video till date yayee have been consistently doing YT and content in the ai space for last 7 months and finally get to feel this😇 i built this tool yesterday called agent router using jev + herdr, i was awake until midnight doing recording, and publishing the video. today it feels all the efforts are worth it

Put together a quick video of using Jev with ReqLLM I cover the new `evaluate/4` method, why I went that route, and make a real API call to Jev to classify an issue https://t.co/wvSoeI07AD

IRS O*NET job classification using Jev (1,016 possibilities) Query: "I scoop scoops and sprinkle sprinkles" Result: 35-3023.00 Fast Food and Counter Workers https://t.co/SSXUqQwNo3

I'm using Jev (from @typesafeai) to categorize expenses for my company! Our office manager used to do this. It's all in Mongolian and we have to do it using bank records. It is very unique to our company, and not something a software would handle easily. We also have about a years worth of Excel files (training data) that I'm using to give Jev guidance on classification (few shot style). It really works, it's stupidly cheap, and when it's not confident it says so. Pretty great! This isn't anything new, classification in ML is extremely "solved". But this is a general classification model