Guillaume Laforge

Fixing #AISlop with a #SKILL.md in #geminicli

Well, at least, removing as much sloppiness as possible from AI generated text! And making it more human-readable ☺️

Inspired by the awesome tropes.fyi

Explanations and skill:
glaforge.dev/posts/2026/0...

I'm not using AI to replicate artist works. I'm using it to discover new vocabulary in a foreign language.
Indeed it reused a familiar character for illustrating the vocab, and yes, there are some small text / ref errors, but the vocab is there and correct, AFAIK.

Not being a native English speaker I miss some vocabulary.

I wanted to know the words describing the elements of a comic book.

My friend #NanoBanana came to the rescue! I asked:

"Create an illustration describing the various elements of a comic page (panels, bubbles, etc)"

TamboUI is inspired from RataTUI.
In French, TamboUI sounds like "Tambouille" which is about cooking, and RataTUI sounds like "Ratatouille" which is a dish.
So the names fit well together!

Smarter Filters: Empowering Users with AI-Driven Search Discover how AI is revolutionizing search filters! Learn how to transform natural language queries into structured filters using Chrome's Built-in Prompt API and Gemini Nano. Explore techniques like s...

I am a bit late sharing this! 😅

I wrote a post a few months ago about building a client-side AI agent. It uses the Chrome Prompt API with structured output to turn natural language into precise search filters.

💸 Zero server costs
🔒 Total privacy
📶 Offline support

bandarra.me/posts/ai-sma...

Detecting and Editing Visual Objects with Gemini | Towards Data Science A practical guide to identifying, restoring, and transforming elements within your images

Interested in extending your image-processing toolkit? Laurent Picard offers a comprehensive guide to identifying, restoring, and transforming elements within your image using Gemini.

Updated my Prompt API demo! 🚀

Now using WebMCP to expose on-page agents to broader browser agents. This lets a general agent recruit the page as a domain-expert "sub-agent" for hyper-accurate tasks.

Links in replies! 👇

#WebMCP #BuiltInAI

🍌⚡️ My favorite image model, now at flash speed!

#NanoBanana 2, pretty much the same quality as the pro version, but super, super fast!

blog.google/innovation-a...

Starting now!

📢 Tomorrow at 🕐 1PM (Paris) / 🕛 noon (London), I'll be your host for our #WednesdayBuildHour

I'll guide you through an online workshop to learn how to
𝗕𝘂𝗶𝗹𝗱 #AI #Agents 𝘄𝗶𝘁𝗵 #ADK 𝗳𝗼𝗿 #Java

➡️ Register here: goo.gle/wbh-adk-java

Easily Build a Local MCP Server in Java with a Skill inside Gemini CLI Recently, I’ve been exploring the Model Context Protocol (MCP) and how to easily create custom servers to extend the capabilities of AI assistants like Gemini CLI which I use daily. I wanted a way to ...

#Agent #skills help 𝗮𝘂𝘁𝗼𝗺𝗮𝘁𝗲 𝗿𝗲𝗽𝗲𝘁𝗶𝘁𝗶𝘃𝗲 𝘁𝗮𝘀𝗸𝘀.
Recently I used #GeminiCLI to easily create #MCP servers in #Java with @jbang.dev & @langchain4j.dev.
I extracted that 𝗸𝗻𝗼𝘄𝗹𝗲𝗱𝗴𝗲 into a 𝗿𝗲𝘂𝘀𝗮𝗯𝗹𝗲 SKILL.md to create more servers!
glaforge.dev/posts/2026/0...

Is your personal #AI #agent following more the 🦞 #lobster philosophy or the 🐜 #ant philosophy?
(like in the #openclaw / #ironclaw / #microclaw / #nanoclaw / #picoclaw... community)

📈All those people showing graphs with projected exponential growth about something... but future turns often more into a sigmoid function!

👋 Hello from the plateau!

Comparing #Lean and #Rocq (Coq)
I thought both were theorem provers, but it seems Rocq is more for building verifiable software, while Lean is for more classical math.
So Lean might be a better fit for the math explorations #LLMs and #AI #Agents are doing for solving problems.

Google's Gemini 3.1 Pro is mostly great Gemini 3.1 Pro tops most benchmarks at half the price of Anthropic's Opus 4.6.

Gemini 3.1 Pro tops most benchmarks at half the price of Anthropic's Opus 4.6.

Image

Introducing Gemini 3.1 Pro, our new SOTA model across most reasoning, coding, and stem use cases! x.com/OfficialLoga...

Java CLIs in 2026:

- scaffold with PicoCLI picocli.info
- style with Tamboui tamboui.dev
- launch with JBang www.jbang.dev
- release with JReleaser jreleaser.org

Sometimes, the #AI talks to me...

I built a research workflow that uses #Gemini's interactions API, Google's Deep Research agent, also #nanobanana for illustration, and due to a timeout in my system, the sketchnote agent was yelling at me because I forgot to feed it! 🍴😋

Perso, le cours de compilation pendant les études n'était vraiment pas mon cours préféré... et pourtant j'ai passé tant d'années sur... un langage de programmation 😉

Gemini Deep Think: Redefining the Future of Scientific Research Gemini Deep Think is accelerating discovery in maths, physics, and computer science by acting as a powerful scientific companion for researchers.

Aletheia (basé sur Gemini Deep Think) a aidé des chercheurs à résoudre 18 problèmes bloqués depuis des années en apportant des approches mathématiques inattendues. Dont une conjecture vieille de 10 ans démolie par un contre-exemple.

📣 Woohoo!
A proper real #TUI library for #java!!!! Big Day!
Congrats @melix.champeau.me and @maxandersen.xam.dk 🥳

Text-to-speech generation (TTS)  |  Gemini API  |  Google AI for Developers Get started generating audio with the Gemini API

You can see the various voice options here:
ai.google.dev/gemini-api/d...
I guess I could add an enum for them, but as they change or some or added, I'm not sure I'd be able to properly keep track.

I released v0.7.1 of my #Gemini Interactions API Java SDK adding the ability to do speech-to-text and customizing voices
github.com/glaforge/gem...

But don't get me wrong, Go is also a really great option 😉

High-Performance AI on Cloud Run: Deploying Gemma 3 with RTX 6000 Pro GPUs Cloud Run now offers (in preview) high-end inference with support for NVIDIA RTX PRO™ 6000 Blackwell Server Edition GPUs on Cloud Run…

High-Performance #AI on #CloudRun: Deploying #Gemma3 with #nvidia RTX 6000 Pro GPUs
medium.com/google-cloud...

They haven't tested Java with @graalvm.org native compilation. Or with more efficient frameworks like @quarkus.io or @micronautfw.bsky.social. Then Go would likely lose its advantage on memory.

⚠️Si jamais vous tombez sur cette boutique en ligne, #Dundle, passez votre chemin.
💩Ce sont des arnaqueurs!
Ma fille a essayé d'acheter plusieurs "game cards" pour des jeux vidéo pour un anniv.
Aucune ne marchait.
Le site ne veut évidemment pas rembourser...
👺ARNAQUE 👹

Java Conferences with closing CFP in the next 15 days:

JJUG CCC 2026 Spring Closes: 1 March 2026
DevBcn Closes: 28 February 2026
Devoxx Poland Closes: 28 February 2026

If you want to add your conference, please submit it at https://javaconferences.org/

Interesting... #Java is actually a great language choice to implement your #MCP servers, if you care about performance!

And if compiled natively with @graalvm.org the memory would be as low as #golang for similar performance!

www.tmdevlab.com/mcp-server-p...

In graph theory, there are algorithms that find the shortest path between two nodes. I made one with pure CSS (including the graph drawing).

Drag the nodes, and the shortest path will update in real-time!

css-tip.com/graph-theory/

A fun demo powered by all the modern & cool CSS features 🤩