Guillaume Laforge

There's even a #darkmode 😉

You can try it online:
adk-agent-code-visualizer-1029513523185.europe-west1.run.app

And have a look at the code:
github.com/glaforge/adk...

⚠️ It works for both single-file #Java and #Python ADK multiagents. It doesn't work for projects spanning multiple files or directories.

✨ Vibe-coded with Google #AiStudio and #GeminiCLI.
Powered by #Gemini 2.5 Pro.
Using #ReactFlow for the nice node-based UI.
Deployed on #CloudRun ☁️

Visualizing ADK multiagent systems Let me share an interesting experiment I worked on to visualize your AI agent structure, more specifically, Agent Development Kit (ADK) multiagents. The more complex your agents become, as you split t...

A proof-of-concept #ADK #AI #Agent code visualizer
glaforge.dev/posts/2025/0...

YouTube video by Google Cloud Nuggets Build AI Agents in Java A Guide to the ADK

For my series of articles on #ADK for #Java, my colleague Romin Irani used the new "Video Overview" feature of #NotebookLM to generate this great overview of my articles! It's impressive!

www.youtube.com/watch?v=HqNS...

The recap of the series can be found here: glaforge.dev/posts/2025/0...

Mastering agentic workflows with ADK: the recap Over the past few articles, we’ve taken a deep dive into the powerful agentic workflow orchestration capabilities of the Agent Development Kit (ADK) for Java. We’ve seen how to build robust, specializ...

🔚 And here is the final recap article on the various #ADK agentic workflows on when & how to use

1⃣ sub-agents
2⃣ sequential
3⃣ parallel
4⃣ loop agents

And their typical use cases, ✅ pros & ❌ cons.

What would you like to learn about, next?

glaforge.dev/posts/2025/0...

Mastering agentic workflows with ADK: Loop agents Welcome to the final installment of our series on mastering agentic workflows with the ADK for Java. We’ve covered a lot of ground: Part 1: Sub-agents for flexible, user-driven delegation. Part 2: Se...

👓 Read all the details about #ADK #Java loop flows for your #AI #agents in this article ⬇️

glaforge.dev/posts/2025/0...

The last of the series on agentic workflows! 🔚

🔁 What about〖 loop flows 〗with #ADK for #Java for refinement, trial/error, self corrective #AI #agents?

We'll talk about ⏪ before & ⏩ after agent callbacks, function calling exit, and max iteration limits ♾

Concrete example: a simple #Python code refinement loop agent 🧵

And stay tuned, as I still have to tell you about the "loop flow", where you can make several agents work in a loop, until some condition is satisfied.

Mastering agentic workflows with ADK for Java: Parallel agents Let’s continue our exploration of ADK for Java (Agent Development Kit for building AI agents). In this series, we’ve explored two fundamental agentic workflows: First, we used LlmAgent with sub-agent...

This #AI #Agent actually combines parallel running agents, as well as a sequential flow with a final agent compiling all the research materials.

Combining different kind of flows makes sense for complex scenarios.

Read all the details in this new article:
glaforge.dev/posts/2025/0...

📢 Today let's talk about〖 parallel flows 〗with #ADK for #Java. Several agents can run at the same time when their tasks are unrelated.

A 🏭 company researcher agent:
1⃣ a company profiler
2⃣ a news finder
3⃣ a financial analyst run in parallel

📢 After yesterday's post on #ADK〖 sub-agents 〗in #java, let's have a look at the〖 sequential flow 〗of agents.

🗺️ With an example of a trip planner, with an #AI #agent searching info about a destination, an itinerary agent, and a restaurant finder.

glaforge.dev/posts/2025/0...

Mastering agentic workflows with ADK for Java: Sub-Agents Let me come back to the Agent Development Kit (ADK) for Java! We recently discussed the many ways to expand ADK agents with tools. But today, I want to explore the multi-agentic capabilities of ADK, b...

New article exploring the various agentic workflow patterns in #ADK for #Java. This time, zooming in on "sub-agents".

Next we'll explore sequential, parallel and loop flows.

glaforge.dev/posts/2025/0...

The Sci-Fi naming problem: Are LLMs less creative than we think? Like many developers, I’ve been exploring the creative potential of Large Language Models (LLMs). At the beginning of the year, I crafted a project to build an AI agent that could generate short scien...

Read the full story here on the apparent lack of creativity of #LLMs
glaforge.dev/posts/2025/0...

I also used #GeminiCLI to search through those datasets!

So I wondered where those names were coming from, and my intuition was that LLMs were drawing their inspiration from a limited set of sci-fi stories with little naming diversity.
I searched on #Kaggle and found a couple datasets where those names appeared often.

In my AI agent generating sci-fi stories (developed with
@langchain4j.dev, deployed on #CloudRun) I was always encountering the same names again and again:
short-ai-story.web.app

The Sci-Fi naming problem: Are #LLMs less creative than we think?
The most well known LLMs seem to always give the same names to the protagonists of #scifi stories.
Why do they do that? Are they not creative?
Who are Dr Thorne and Anya?

À la rentrée, je serai à Tours, à @tadx.bsky.social, pour parler agents IA ! 🤖🧠

🧵 As an (Gen)AI experiment, I kicked off a new project in React (which I had zero experience with) using Vibe Coding. The first few days were incredibly productive, and within a week I had 90% of the project done. But for that last 10%, Claude Sonnet 4 just couldn’t help me anymore...

Just shared my presentation on #AI #Agents.

If you want to learn more about the #MCP & #A2A protocols, and frameworks like #ADK, @langchain4j.dev for building agents in #Java in particular, read on!

glaforge.dev/talks/2025/0...

Using #Gemini and long context for indexing rich documents (PDF, HTML... containing images & diagrams) for your #RAG pipelines
glaforge.dev/posts/2025/0...

Tranches de Tech 20 - Quand Java rencontre l'IA Présentation de Guillaume https://glaforge.dev/  https://lescastcodeurs.com/  https://javachampions.org/ https://www.groovy-lang.org/ https://docs.langchain4j.dev/ https://github.com/tmc/langchaingo  ...

Le dernier épisode de Tranches de Tech est disponible : smartlink.ausha.co/tranches-de-... 🎧

Merci @glaforge.dev pour ta disponibilité et la qualité de nos échanges 🤩.

Merci #OVHcloud de nous permettre de continuer cette aventure ❤️.

On se retrouve en septembre 👋.

Bonnes vacances 🏝️ ☀️.

I hope you will finish your draft!
I'd be happy to read it!

You mention the hybrid search semantic+BM25 but there's another kind of "hybrid" that you could add as a new bullet point in your draft: RAG+large context.
You find relevant vectors, but you feed the complete doc in the large context window of the model. It solves the problem of the enterprise plan.

I like when you say "the embedding is diluted". I often use that adjective to intuitively describe what happens with bigger chunks!

Advanced RAG — Hypothetical Question Embedding In the first article of this Advanced RAG series, I talked about an approach I called sentence window retrieval, where we calculate vector embeddings per sentence, but the chunk of text returned (and ...

I vibe-coded an application to test this approach with #Gemini Canvas and #GeminiCLI, and I deployed it on #CloudRun.

You'll find the link and the implementation details with @langchain4j.dev in my article.

glaforge.dev/posts/2025/0...

Rag Strategies Hypothetical Questions and HyDE | PIXION Blog Exploring Hypothetical Questions and HyDE strategies, designed to boost the accuracy of LLM-based applications by refining the match between user queries and data. It examines the generation of hypothetical questions and answers emphasizing the contrast in their approach.

The #LLM is responsible for figuring out which questions can be answered by the chunk.

This is a technique I discovered on this blog:
pixion.co/blog/rag-str...

(along with the explanation of another approach called HyDE, for Hypothetical Document Embedding)

Just wrote a new article to explore a #RAG ingestion technique that I like to use for Q&A oriented apps:

💡Hypothetical Question Embedding

The idea is to compare user questions to #LLM generated questions extracted from the chunks of text.