Maxime Ferret

Software engineer (8 years)
Love AI, building things, internet businesses

And a lot of other things

[Deedy @deedydas]

You can now generate an entire 3blue1brown style video from any research paper with Opus 5.5.

Here’s a 8min video summary of “Regularized Recursive Self Improvement of Agent Harnesses”.

The 90%ile educational YouTuber is fully automated.

[liam. @liambraus]

ESTO ES JODIDAMENTE INCREÍBLE

un dev en solitario acaba de abrir el código de un reemplazo 100% GRATIS de ElevenLabs que se ejecuta completamente en tu propia máquina.

el repo de GitHub está en 19.4K estrellas.

te permite:

→ clonar una voz desde un clip de referencia limpio
→ doblar cualquier video a 646 idiomas
→ generar audiolibros, dictado, transcripción
→ elegir entre 14 motores TTS en lugar de uno

ElevenLabs soporta 32 idiomas. esto hace 646.

sin cobro por carácter. sin límites de uso. ningún audio sale nunca de tu computadora.

guarda esto para después.

os dejo el repo abajo

[StudioYebisu @studio_yebisu]

やぁ！兄弟達！

一晩寝ただけで、Jev系のGitHubリポジトリはめっちゃ増えたから!!
ジャンル別にスターが多かったものをまとめておいたよ!!
スターの数はリサーチ時点のものだからそこは勘弁してほしいぜ!! ハハッ

Metaが出してるCV。SAM3.1もAPIで使えるようになったってのと。
JevもOpenRouterで使えるようになったみたいだ!!
もうウェイトリストに並ぶ必要はないよ。

CV x Jevで何か面白いものを作ってよ!!
今日も今日とてGrok翻訳みたいな文章はわざわざタイピングしてるヨ!!

【ブラウザ・PC・スマホ操作】
1. jev-ultrafast｜★4,891
圧倒的大人気！JevといえばUse系への活用が一番期待されてるみたいだね!!
ブラウザ操作と対象要素の選択をJevが担当し、テキスト入力時のみ小型LLMを呼び出す構成。
https://t.co/aZfA1mfOHX

2. typesafe-computer-use｜★203
macOS画面のOCRやアクセシビリティ情報を読み取り、クリックや文字入力を自動化するMac専用ツール。
https://t.co/nxPWCeWFPk

3. mobile-jev｜★103
Mobilerun接続のAndroid端末を操作。CLIやWeb画面、実行ログを確認しながら操作を進められる。
https://t.co/JKGDRx8gYV

4. jev-voice-browser｜★40
☆これ思ったより面白そうだったよ!!☆
ブラウザの音声認識から意図を抽出し、Playwright経由でWebページを操作する実験的実装。
https://t.co/21h1dM8du1

【AI開発・コードレビュー】
5. fast-jev-compaction｜★2,790
Claude Codeの履歴から、不要と判断したツール呼び出しや結果を削除・短縮。残す会話文は原文のまま。
https://t.co/cOuB4E7xcb
※これも、めっちゃ人気だね！Claudeの不要な部分をJevでコントロールするのは良いユースケースみたいだ!!

6. foreman｜★279
Codexの自律作業をJevが監督し、作業の継続や検証、停止を判断する開発プロセスの実験。
https://t.co/8BuweE6c1j
Codex版だね！

7. jev-review｜★251
Gitの差分やリポジトリ全体を段階的にレビューし、潜在リスクやテスト不足をローカル画面に表示。
https://t.co/0N2PRprSws

8. jev-router｜★121
Claude CodeやCodexの各ターンで、要求タスクの性質に合わせて呼び出すモデルを自動で振り分ける。
https://t.co/I0uCLYk8jI

9. jev-rules｜★3（新顔）
依頼内容や編集対象ファイルに応じて、Claude Codeへ渡す指示ルールや参照資料を選ぶ。
https://t.co/YTP8pxMGw5

【MCP・スキル・CLI】

10. skillbox｜★153
エージェント用スキルを管理・MCP配信する基盤。追加のJev連携で、タスクに合うスキルを推薦する。
https://t.co/PKwPfJaued
これもまるでハーネスの一部として、Jevを活用するってのは面白いよね！！

11. typesafe-mcp｜★62
ClaudeやCodexなどのMCP対応環境から、Jevの選択・採点・真偽判定を直接呼べる非公式ツール。
https://t.co/CoU8h9Dcb1

12. jev-shell-history｜★30
zshで入力中の内容に合う過去のコマンドをJevが選び、履歴から補完する。
https://t.co/jSyj38q394

【検索・データベース】
※このユースケースは注目だ！
Jevの評価機能を上手く使って、ソートをハックしてく。

ユーザーの行動にリアルタイムで合わせて選択をフィルタリングしていくってのはWebと相性が良さそうじゃないか？

13. pg-jev｜★143
PostgreSQLの行を、SQL内の自然言語条件で絞り込み・分類・順位付け。対象データは外部APIへ送信。
https://t.co/PDcAQiOxiV

14. jev-search｜★48
検索語の選定や対象期間の指定、取得したWeb検索結果の関連度ソートをJevに任せる検索パイプライン。
https://t.co/FO6xvZO4lr

15. neo4jev｜★17
Neo4jのグラフ構造において、次にたどる関係をJevに選ばせ、データのつながりを探索・可視化するデモ。
https://t.co/uoeOzKuCVe

【動画・Web閲覧】

16. unclutter｜★73
Webページ内の広告枠や邪魔なポップアップをJevで識別し、画面から非表示化するブラウザ拡張。
https://t.co/kwJsPBb0Ne

17. jevmeter｜★52
動画内の発言を指定した観点で採点し、メーター付きの動画を作る。事実確認や嘘の判定ではない。
https://t.co/nZzlODHtRg

18. youtube-sponsor-detection｜★46
YouTubeの字幕や音声から動画内のスポンサー紹介区間を判定して自動スキップ（配信広告除去とは別）。
https://t.co/iAgNa0Vq5G
※人類はどうしても広告を見たくないみたいだね！ハハッ！

【ガードレール・コミュニティ管理】
19. pi-warden｜★61
Piエージェントを監視し、プロジェクトのルール違反、同じ失敗の繰り返し、未検証の完了報告をチェック。
https://t.co/dbHkInBlxq

20. Jev-Moderation-Bot｜★25
Discordのスパムや詐欺URLを判定し、メッセージ削除や警告、タイムアウトを管理する。誤判定への管理者対応は必要。
https://t.co/hWvzIY1QYz

【マーケティング】
21. notra｜★171
AI回答でのブランド言及を追跡するGEOツール。言及の好意度や掲載順位などの分類処理にJevを活用。
https://t.co/vu1oa7FkJW
※おっと、こいつも注目だ！
GEOツールとして使うことで、事前にA/Bテストをしておくみたいなもんだな。Jevの評価の早さを上手く活かしてるぜ！

【スマートホーム】
22. HA-Jev｜★6（新顔）
Home Assistantの家の状態をJevで判定。洗濯物通知など日常の自動化向けで、安全設備には使わない。
https://t.co/3Mk680U0ta
※Homeへの実装は進めてほしいよな！

【ゲーム・ロボットのシミュレーション】
23. typesafe-mario｜★260
エミュレーターから抽出したゲーム状態データをもとに、Jevにマリオの行動を選択させる制御実験。
https://t.co/r2nQgGACcH

24. jevpilot｜★58
Three.js製の走行シミュレーター上で、提示された進路と速度の候補からJevが行動を選択する実験。
https://t.co/bVJ9HdUM1Y

25. jev-drone｜★58
物理エンジンMuJoCo上のドローンシミュレーションで、障害物の回避行動の判断をJevに委ねる実験。
https://t.co/fmLUYdDSWD

【売買ボットの実験】
26. jev-trader｜★804
Monad上のKuruで売買判断を試す実験。既定は模擬判断のmock、鍵なしは模擬取引のdry-run。収益性は未検証。
https://t.co/DODHIGpTJi
※これも注目度が高い。だれかテクニカル分析の専門家は自分の環境でチューニングしてみてくれよな。

【ローカル・オープンモデル研究（公式Jevとは別）】
以下4件は独立した研究・実装。公式Jevの公開版ではありません。
27. SemIf｜★1,500
公開モデルで選択肢の確率を直接読み出す独立研究（旧OpenJev）。ブラウザ上で動くデモも提供。
https://t.co/q2OZKmInGE

28. jevlike｜★851
可変長の選択肢群を一度に評価する小型モデル学習の独立研究。公式Jevの複製ではなく研究の出発点。
https://t.co/Ke9wmLHCTU

29. NanoJev｜★313
Qwen3-0.6Bベースの並列判断モデル。重みや学習コードを公開し、ゲーム制御デモなどを同梱した独立実装。
https://t.co/cQCR6VytW1

30. openjev-sglang｜★127
QwenとSGLangによる互換APIの独立実装。B200などGPUサーバーでの稼働を想定した構成。
https://t.co/QXklNwzA79

おっと、最後に注意事項だ。
API課金・外部へのデータ送信���使う前に各リポジトリで確認を。今回は公開資料を調べた紹介で、全件を試用・動作検証したわけではありません。

ある程度は内容を目視でみたけどな！流石に全部を検証する時間はなかったぜ！

[@levelsio @levelsio]

When I set up a new Hetzner VPS first thing I do install Tailscale and once I'm in via Tailscale lock down the firewall to only accept web traffic on HTTPS 443 for Cloudflare IPs and SSH 22 for Tailscale IP

That way nobody can get in

I know I keep repeating this but it should be basics of setting up a new VPS

So basic IMHO it should be part of any VPS service to default install Tailscale and enable it so it's the only way to get in

Why?

A VPS server is just like your laptop or destop computer but now imagine if it's connected to the entire internet with 8 billion people that can access it and try hack it

You want to only have it accessible to you

And if you want to host a website on your VPS (like I do), you should only let Cloudflare access your VPS so it can stand in front and block any hack attempts

Never expose a VPS to the world wide web which realistically is the world WILD web

[Mike Futia @mikefutia]

Claude Code + Nano Banana 2 is f*cking cracked 🤯

I built a system inside Claude Code that researches any brand, writes 40 ad prompts from scratch, and fires them all to Nano Banana 2.

One brand name + one URL = 40 production-ready static ads.

All inside Claude Code.

I took @alexgoughcooper's brilliant framework and automated the whole thing inside Claude Code.

Perfect for DTC brands and agencies who need high-volume ad creative without briefing a designer or spending hours in Canva.

If you're finding winning ad concepts on Meta and manually recreating them one at a time in Higgsfield — copying prompts, pasting product details, tweaking aspect ratios, downloading, organizing...

This system eliminates the entire loop:

→ Give Claude a brand name and URL
→ It researches the brand's fonts, colors, packaging, and photography style
→ Builds a Brand DNA document from scratch
→ Fills in Alex's 40 proven ad templates (headline, us vs them, testimonial, UGC, review cards, stat callouts) with brand-specific details
→ Fires every prompt to Nano Banana 2 with your product photos as reference
→ Downloads finished ads into organized folders with an HTML gallery

No Higgsfield.
No manual prompt filling.
No copy-pasting between tools.

What you get:

→ 40 ad formats filled with your exact brand colors, fonts, and copy
→ 4 variations per format so you pick the best output
→ Product photos passed as reference so the model matches your real packaging
→ A reusable system — new brand, new folder, same pipeline

Built 100% in Claude Code with Nano Banana 2.

I put together a full playbook & Loom video showing the exact process to set this up yourself.

Want access for free?

> Like this post
> Comment "NANO"

And I'll send it over (must be following so I can DM)

[Min Choi @minchoi]

Ok Clawdbot is insane.

People can't stop building and founders are making money with it.

10 wild examples + how to setup (No-code required guide in ~30 minutes).

Bookmark this👇

[Min Choi @minchoi]

Nano Banana Pro just cooked ChatGPT Image 1.5

Insane brand ads, product concepts, and multi-angle shots just from images & text.

10 wild examples

[Min Choi @minchoi]

Ok Claude Code is insane.

People can't stop building and saving time & money with it.

10 wild examples.

[Yasser @yasser_elsaid_]

I scaled Chatbase from a side project to a $6M ARR startup. No sales team, no VCs, just product‑led growth.

Here is the full strategy for scaling to millions purely through product-led growth.

[MaximeB @Buildingitmyway]

Traffic for my app keeps increasing exponentially on autopilot with pSEO. 📈

Every 3 hours a new page is indexed by Google, driving long-term profitable organic traffic. 🌱

Here's how you can do it too, full guide 👇

[@levelsio @levelsio]

It's so exciting to be in tech right now

He's 100% right, it's like the early days of the internet around 1995 again

Everything was new and you had endless opportunities, you could build the first web browser, or mail app, or video streaming tech, etc

But instead of internet, it's now the early days of AI and 1995 is 2025

Most people don't realize it but this is a VERY special time with SO many opportunities right now, all you need is to build

And times like these only happen every ~30 years if at all, jump on it!

[Alex Albert @alexalbert__]

I just connected Claude to an internet search engine using MCP.

Here's how you can do it too in under 5 minutes:

[CJ Zafir @cjzafir]

AI coding = Proper Documentation + Detailed Instructions

It's like a smart kid who can code but doesn't know the best practices of code.

You need to tell AI what exactly you want:
- web app?
- features?
- structure?
etc

Here's how you can make AI your full stack developer:
🧵

[AshutoshShrivastava @ai_for_success]

Cursor is rolling out Agents.
It can now pick its own context, use the terminal, and complete entire tasks. 🔥🔥
What's next for Software developer?? Just open the IDE and chill???

[Akshay 🚀 @akshay_pachaar]

Traditional RAG vs. HyDE, visually explained!

One critical problem with the traditional RAG system is that questions are not semantically similar to their answers.

Consider the example where you want to find a sentence similar to "What is ML?" It is likely that "What is AI?" will appear more similar than "Machine learning is fun."

This semantic dissimilarity leads to several irrelevant contexts being retrieved during the retrieval step.

HyDE solves this issue.

The visual below illustrates how this approach differs from traditional RAG.

Here's how it works:

- Use an LLM to generate a hypothetical answer (H) for the query (Q). This answer doesn't have to be entirely correct.

- Embed the answer using a contriever model to obtain E. Bi-encoders trained with contrastive learning are generally used for this purpose.

- Use the embedding E to query the vector database and retrieve relevant context (C).

- Pass the hypothetical answer H, the retrieved context C, and the query Q to the LLM to produce a final answer.

Done!

Now, of course, the hypothetical generated will likely contain hallucinated details.

But this does not severely affect the performance due to the contriever model—one which embeds.

More specifically, this model is trained using contrastive learning and it also functions as a near-lossless compressor whose task is to filter out the hallucinated details of the fake document.

This produces a vector embedding that is expected to be more similar to the embeddings of actual documents than the question is to the real documents.

Several studies have shown that HyDE improves the retrieval performance compared to the traditional embedding model.

But this comes at the cost of increased latency and more LLM usage.

Enjoyed this? You should also my RAG series! From building and optimizing RAG apps to evaluating performance and crafting agentic & multi-modal systems—it's all here.

Link in the next tweet!
_____
Find me → @akshay_pachaar ✔️
For more insights and tutorials on AI and Machine Learning!

[Moritz Kremb @moritzkremb]

I used the new Cursor agent + CopyCoder to build a Calendly frontend clone in 20mins

Cursor agent now:
- runs terminal commands
- picks it's own context
- executes more jobs with one prompt

Crazy where this is going

[Joschua Sutee @JoschuaBuilds]

350 Places to post your product (🧵):

[Alex Albert @alexalbert__]

Here's a quick demo using the Claude desktop app, where we've configured MCP:

Watch Claude connect directly to GitHub, create a new repo, and make a PR through a simple MCP integration.

Once MCP was set up in Claude desktop, building this integration took less than an hour.

[Maxime Ferret @maximebf_]

@levelsio Do you store both your database and server on the VPS ?

[@levelsio @levelsio]

✨ First AI talking video made from within Photo AI with the new [ 💬 Make talking video ] interface

AI photo -> AI video -> AI lipsync -> AI captions

Obviously not very passable yet, BUT it's a start and with voice and lipsync models improving so fast I think it'll be passable in 3-6 months

One thing @elevenlabs is severely missing is background sounds which I didn't add here manually just to show the state of things, is there any good TTS API that DOES have bg noise?

Captions are by CapCut, but it'd be nice to be able to do that in-app with Photo AI too, not sure where to start with that though

Again cool thing of this vs HeyGen is it has no limits to where you wanna place yourself or your model, the source is just a prompted AI photo in Photo AI that becomes a talking video with 2 clicks