[May 2025] AI Community — Activity Highlights and Achievements

[May 2025] AI Community — Activity Highlights and Achievements

Let’s explore highlights and accomplishments of the Google AI communities over the month. We appreciate all the hard work and dedication of our community members. Here are the key highlights!

Featured stories

Jules: Google’s Codex Killer? by AI GDE Muhammad Farooq (US) is an introducing video of Jules, new asynchronous AI coding agent. Muhammad shares key features and main capabilities of Jules.

Google Jules: A Guide With 3 Practical Examples by Aashi Dutt (India) is a tutorial that explores, step by step, how to connect Jules and GitHub to autonomously handle real-world software tasks using 3 examples. Aashi also shared two more tutorials on Gemini Diffusion and Gemma 3n.

AI Movie Making: Massive steps forward by AI GDE Laurence Moroney (US) shows how he created videos of his novel using Veo 3 and shares tips to achieve consistency in creations and prompts he used.

Sketch2Vid using Gemini & Veo 3 by AI GDE Nitin Tiwari (India) an implementation of converting sketches into lively videos using Veo 3 model.

Radiology with MedGemma & Gemini’s Native TTS by AI GDE Rishiraj Acharya (India) is a Kaggle notebook that features a radiology voice assistant powered by MedGemma, which translates complex medical image reports into simple, understandable language. Combined with Gemini’s TTS, the assistant provides a voice-driven experience making medical insights more accessible.

Googlers on AI GDEs’ YouTube Channels

• Google’s New Stack: Gemini On-Prem, ADK, Open Models by AI GDE Muhammad Farooq (US) is an interview video with Matt Thompson (Director of Developer Advocacy at Google Cloud) delves into ADK highlighting its potential to enable developers to build AI agents through an open-source framework. It also explores Gemini on-prem, support for open weight models like Gemma, and TPUs, and the future of developer tooling, and etc.
• Special Episode with Nate Keating by AI GDE Taha Bouhsine (US) is an interview video with Nate Keating (Head of Kaggle) discussing the role and impact of Kaggle as a platform for AI education. They also talk about the too perfect AI-generated content that may be imperfect to the human learning process. This discussion emphasizes problem-solving skills and developer communities will remain crucial even in the AI landscape.
• Building Bridges with Developers by AI GDE Allen Firstenberg (US) is an interview with Ankur Kotwal (Global Head of Cloud DevRel) dives deep into the world of Developer Relations (DevRel) at Google, discussing its crucial role as a bridge connecting Google’s product teams and engineers with the global developer community. By sharing their personal experiences as a DevRel lead and AI GDE themselves, they emphasize on the importance of community programs, developer communities, and continuous learning, etc.

Product Highlights

Agent Development Kit

Google ADK + Vertex AI Live API (starter code) by AI GDE Sascha Heyer (Germany) is an article about integrating ADK with the Multimodal Live API for real-time audio and video streaming with AI agents. It provides a detailed explanation of the process, including initializing ADK components, sending media with LiveRequestQueue, and handling responses from the agent.

[Gemini][LINEBot] Easy upgrade! ADK implementation guide for converting from Function Call to Agent mode (repository) by AI GDE Evan Lin (Taiwan) discusses upgrading a LINE bot from using LangChain’s Function Calling to the ADK. ADK allows for more streamlined code and enhances the bot’s conversational abilities, making interactions more natural and context-aware. The post provides a detailed guide on converting existing code and emphasizes the improved memory and context handling capabilities of ADK.

Build MCP Agents — using Google ADK || MCP Tutorial — Part 2 (Part 1 | Part 3) by AI GDE Tarun R Jain (India) is the part 2 of MCP tutorial playlist. Tarun guides how to use ADK and implement the code to connect the Airbnb MCP server with ADK, powered by Gemini 2.5 Flash.

AI GDE Asei Sugiyama (Japan) organized and participated in a 3-week hackathon, AI Hackathon starting from KOZA. His team developed an AI assistant (slides) that specializes in analyzing the Okinawa dialect to help communication of the residents and elderly people . They used Firebase Hosting as a frontend app and Vertex AI as a backend, and ADK to build an agent.

Critic’s Cut: Building a Safe, Multilingual Movie Review Agent Using Google ADK + Gemini 2.0 (code) by AI GDE Jigyasa Grover (US) shows how she used ADK, Gemini 2.0 Flash, Google Search Tool, and safety libraries like Presidio Analyzer and Better Profanity to make the movie review agent.

Build With AI — ADK and MCP — GDG Belo Horizonte (video | code | code with solutions) by AI GDE Gabriel Cassimiro (Brazil) was a 3.5-hour online workshop about creating multi-agent systems using ADK. Gabriel also demonstrated how to create an MCP server using Fast API and connect it to the ADK.

https://medium.com/media/d7467e754fa37d086d15a07a53e8e6ed/href

Building and Deploying AI Agents with Google ADK and Vertex AI (video) by Angular GDE Muhammad Ahsan Ayaz (Sweden) at TFUG Islamabad event, introduced ADK and its capabilities, Vertex AI integration, and real-world use cases of AI agents. This group also hosted Building with Google Gemini: Prompt Engineering & Multimodal AI with Googler Abdul Raheem.

Gemini

Building a model with Gemini to predict enemy jungler locations on the LOL minimap (4 hands-on Colab notebooks in the slide deck) by AI GDE Junbum Lee (Korea) was a workshop aimed to build a model that predicts the position of the enemy on the minimap of the game, League of Legends.

https://medium.com/media/d847a3a819ef8843b3a8f51311af2a23/href

AI GDE Chansung Park (Korea) shared demos using Gemini 2.5 Pro: TCP/IP Packet Simulator, Turning diagram into interactive web components (for Nvidia Dynamo), Multi Agentic System Simulator, Neural Weave (Gemini app for Gemini), and Website creation and previewing (in 4 parallel)

From Models to Agentic Applications with Sam Witteveen by Muhammad Farooq (US) is a discussion video on the recent I/O updates and Google’s AI strategies. They share their thoughts and analysis on future trends in AI. Additional videos from the two AI GDEs:
– Gemini 2.5 Pro’s: The Best Coding Model Just got Better by Muhammad
– The Improved Gemini 2.5 Pro — A Coding Powerhouse by Sam
– MedGemma — An Open Doctor Model? by Sam
– Gemini TTS — Native Audio Out by Sam

ACOT — AI Comments Tool for Google Docs by AI GDE Hugo Zanini (Brazil) is a Chrome extension that enhances productivity in Google Docs by using AI to summarize comment threads. It provides instant AI-generated summaries in a convenient side panel.

Learning from Kaggle Competitions using Gemini 2.5: AI Mathematical Olympiad — Progress Prize 2 by AI GDE Luca Massaron (Italy) is an article that explores the competition, AI Mathematical Olympiad — Progress Prize 2 with Gemini 2.5 Pro. Luca examined the available solutions up to the 22nd rank of the private leaderboard and listed those that can help you build effective SLMs capable of mathematical and logical reasoning.

Gemini 2.0 Flash: What can it do?

[200+ 👏] Gemini 2.0 Flash: What can it do? by AI GDE Ifeanyi Idiaye (Nigeria) introduces Gemini 2.0 Flash and explores its capabilities in text generation, image generation/editing, image captioning, understanding video, audio, document, and code. It highlights the model’s potential for developers to integrate AI into their applications.

Building a RAG Chatbot with Google Sheets, Gemini and Lamatic by AI GDE Vrijraj Singh (India) is a codelab that guides you through integrating Google Sheets & Gemini with Lamatic.ai to create a chatbot. It covers how to connect your data, process it through chunking and vectorization, build a RAG pipeline using Lamatic’s visual builder interface, and deploy a functional chatbot that can answer questions based on the content of your spreadsheet.

Make Telegram smarter! Gemini × RAG × n8n full guide by AI GDE Kevin Chiu (Taiwan) was a hands-on workshop building your own telegram assistant agent using Gemini 2.5 Pro and n8n.

https://medium.com/media/a665e499fd1521d399a0dcd8504c09a2/href

Build with AI 2025 (slides & notebooks) by Artificial Intelligence and Machine Learning Malaysia covered from introduction to Gemini & Gen AI to how to use them in Vertex AI and how to automate workflows using RAG, Function Calling, etc.

AI GDE Cyrus Wong (Hong Kong) shared a series of video about Gemini Code Assist: “Unboxing #gemini #codeassist #Enterprise Edition” — Setup, Free Trial, Private Repo, Code Customization. These videos take an in-depth look at Gemini Code Assist Enterprise Edition and demonstrate its core feature: code customization, which allows the AI to analyze private repositories to adapt to a company’s unique coding style. Cyrus provides setup guides and insights into the challenges encountered during the free trial, and the limitations of the monitoring capabilities.

Gemma

Fine tune Gemma 3 on Object Detection by AI GDE Aritra Roy Gosthipaty (India) and Sergio Paniego is a fine-tuned Gemma 3 model focusing on adaptive vision and language understanding. You can try it here and see how it achieves in detecting license plates.

Building an Offline AI Learning Companion: My Journey with Gemma (Colab notebook) by AI GDE Geeta kakrani (India) is a post about why and how she used Gemma to create a personal learning assistant for students.

Implementing Cutting Edge AI within Your Server with Maximum Privacy through Gemma (slides) by AI GDE Ardya Dipta Nandaviri (Indonesia) was a talk about how to deploy and run advanced AI models on your own infrastructure while maintaining data privacy and security.

Exploring the Deep Fusion of Large Language Models and Diffusion Transformers for Text-to-Image Synthesis (code) by AI GDE Sayak Paul (India) and co-authors is a research paper that provides an exploration of design space related to recent advances in text-to-image synthesis, specifically the deep fusion of LLMs and DiTs for multi-modal generation. They used a frozen Gemma 2B as the base LLM for most experiments, and Gemma 2 2B to boost performance.

Function Calling with Hugging Face and Gemma 3, Zero To Hero by AI GDE Arthur Kaza (Congo) explored function calling with Gemma 3 and Hugging Face. Participants built their first function, such as get_weather(), and made a model using it. They also simulated real-time logic by wiring up a mini agent flow.

Keras

[Synapse Project] AI GDE Anshuman Mishra (India) contributed to Keras/KerasHub by adding Mixtral (Kaggle model | KerasHub), Qwen1.5-MoE-A2.7B (Kaggle model | KerasHub), and Qwen 3. The PRs have successfully merged.

[Synapse Project] AI GDE Harshal Janjani (UAE) contributed to Keras/KerasHub by adding Moonshine ASR (Kaggle model | Model directory in keras-hub:master). This PR has successfully merged. He also contributed to the Keras 3.9.0 release by raising 14 PRs, 11 of which merged.

Diffusion Models in Action: Visual Generative AI with Keras and KerasHub (slides | code) by AI GDE Muhammad Ghifary (Indonesia) was a workshop delved into the specifics of visual Gen AI, including the evolution from VAEs and GANs to diffusion models. He covered from the fundamental principles of diffusion models to practical aspects including the implementation of diffusion models with Keras and how to utilize KerasHub to access pretrained Stable Diffusion.

Firebase Studio & Genkit

Quick Gemini in React apps using Firebase Vertex AI by AI GDE Anubhav Singh (India) a custom hosted codelab on how to integrate Gemini in existing and new React apps using Firebase Vertex AI in a few easy steps. He also shared 4 other codelabs using Gemini and/or Vertex AI.

Getting Started with Firebase Studio by AI GDE Vrijraj Singh (India) is a codelab that guides you through building a sample recipe generator app using React and Firebase Studio. Vrijraj also shared another codelab, Google Gen AI SDK in Web, which is about how to integrate the Gen AI SDK into your web app.

https://medium.com/media/d9357145834d21e666b2cf34fb4deb32/href

genkit-mcp-client-blender by AI GDE Anubhav Singh (India) is a client application for interacting with Blender through MCP. It allows users to create and manipulate 3D scenes by selecting shapes, colors, textures, and patterns, and providing scene descriptions. The project uses Firebase Genkit for accessing Gemini and as the underlying MCP client.

Create and Deploy a Login/Register Page in Minutes Using Firebase Studio — No Coding Required

The Fast Lane to Full-Stack Apps with Firebase Studio (related post | demo video) by AI GDE Geeta kakrani (India) was a talk sharing how to build and deploy full-stack apps faster than ever with Firebase Studio by designing backend visually, integrate Gemini directly into an app, and seamlessly deploy.

JAX

Optimizing Data Loading Performance in JAX with jax-dataloader and Grain by AI GDE Kartikey Rawat (India) is an article about solving the data loading bottleneck in JAX. It introduces two main solutions: jax-dataloader and Grain and compares their features, performance, and ideal use cases, helping JAX users choose the best data loading solution for their specific project needs.

Sentiment Analysis with JAX: Building a Transformer-Based Text Classifier by AI GDE Taha Bouhsine (US) is a tutorial on training a text classifier using the Flax NNX library on Kaggle TPU. He also shared a video tutorial about How to train your miniGPT with Flax nnx on Kaggle TPU.

Compute Goes Brrr Using JAX and NNX at AI & ML Malaysia by AI GDE David Cardozo (Canada) explored how to leverage TPUs and the JAX library to efficiently train LLMs using NNX.

Cloud & VertexAI

AI GDE Sachin Kumar (Qatar) led a series of workshops Deep dive into the Gemini 2.0 and 2.5 models and applications, Building GenAI apps using GenAI SDK / VertexAI for 3 consecutive weeks covering from introducing Gemini models to building a Gemini-powered app in AI Studio/VertexAI/Firebase Studio.

https://medium.com/media/6b34cf4767e09c7dbe8bb47be5c0f950/href

Several Google Cloud Roadshow events took place in various cities in Indonesia including Semarang, Bandung, Surabaya, Palembang, Bogor, etc. AI GDEs in Indonesia contributed to the events as a speaker.

• Project Management Using Gemini by AI GDE Ibnu S Wardy
• Database for AI Application using Big Query to Manage Your Massive Datasets Efficiently by AI GDE Joan Santoso
• Build a Vision Transformer-based image classifier with Keras by AI GDE Muhammad Ghifary
• Optimize Cloud Cost for Production-Ready Retrieval AI Infrastructure using Firestore and Vector Store by AI GDE Surahutomo Aziz Pradana
• From Query to Action: Creating a Gemini-Powered Task Automation Agent by Ardya Dipta Nandaviri
• Mitigating Security Vulnerabilities on Google Cloud and AI by AI GDE Esther Irawati Setiawan

[May 2025] AI Community — Activity Highlights and Achievements was originally published in Google Developer Experts on Medium, where people are continuing the conversation by highlighting and responding to this story.