![[jun-2025]-ai-community — activity-highlights-and-achievements](https://prodsens.live/wp-content/uploads/2025/07/37151-jun-2025-ai-community-activity-highlights-and-achievements-550x309.png)
[Jun 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
AI Training Campaigns Summary | 1H2025
AI Study Jams | 2 regional initiatives and 128 events organized by 54 communities
H1’25 Korea Cloud AI Study Jam trained 8.3k developers with 36.5k skill badges on Gemini and Vertex AI. ANZ Study Jam trained 1.7k developers with 4.1k completed labs on Generative AI. Playlists from TFUG Islamabad and GDG Ahlen.
AI Math Clubs | 47 events by 10 communities
GDG Ahlen completed a 4-month-long initiative for AI Math Clubs, resulting in 14 videos. ML Nashik, Machine Learning, AI, Deep Learning & NLP Community — Bangladesh, Team Technojam (blog post), TFUG Islamabad also successfully finished their own AI Math journey.
AI Paper Reading Clubs | 50 events by 15 communities
Developer Society hosted 7 paper review events and published 9 blog posts and 1 article about their experiences. Playlists from Machine Learning Uyo, TFUG Islamabad, Machine Learning, AI, Deep Learning & NLP Community — Bangladesh contain a series of videos of their own events!
Kaggle Community Olympiad 2025 | 7 competitions
This year’s initiatives have expanded to tech talks and/or training sessions, such as the university course or virtual workshops.
Developer’s Journey
Anna Muzykina, originally from Ukraine and now based in Germany, is the organizer of GDG Ahlen and a WTM Ambassador. She is deeply passionate about knowledge sharing and fostering community. Since starting GDG Ahlen in Nov 2024, Anna has delivered numerous talks as a Flutter developer and organized over 50 events, including the AI Math Club, AI Study Jams, and the Build with AI series. She actively connects developers and AI experts, not just in her region but also across the globe. (playlist)
Product Highlights
ADK
From Models to Agents: Shipping Enterprise AI Faster with Google’s MCP Toolbox & Agent Development Kit by AI GDE Rabimba Karanjai (US) is an article discussing Google’s MCP Toolbox and ADK for streamlining enterprise AI agent development. It includes code examples and deployment tips using Cloud Run and GKE.
A Guide of ADK, Vertex AI & GenKit for Generative AI Professionals (Español version) by AI GDE Ismael Chaile (Spain) is a guide focusing on the use of ADK, Vertex AI, and GenKit for professionals working in the Gen AI field. The document provides information on leveraging these tools for various Gen AI tasks and applications.
Creating Your Own Agentic Newsletter
[57+ 👏] Creating Your Own Agentic Newsletter (code) by AI GDE Ertuğrul Demir (Türkiye) is a project creating a multi-agent system using ADK and Gemini to generate personalized daily or weekly newsletters based on user interests from Reddit.Deep Dive into the Google Agent Development Kit (ADK): Features and Code Examples by AI GDE Rabimba Karanjai (US) article covers configuring the LlmAgent, extending agent capabilities with tools (function calls, code execution, RAG), orchestrating agent workflows, planning and reasoning strategies, managing state and memory, and advanced customization with callbacks.
Automation Order to Cash in SAP with Google Gemini and Agent Development Kit ADK (code) by AI GDE Felipe Lujan (Canada) is a multi-agent system built with ADK that manages the complete Order to Cash business process in SAP environments using SAP OData APIs through MCP.
Roast My Resume (demo) by AI GDE Regnard Raquedan (Canada) is a multi-agent system that analyzes resumes and job descriptions to provide a glow-up plan with actionable advice to stand out and land jobs. Regnard used ADK and Gemini to build it.
Wingspan Agent by AI GDE Vikram Tiwari (US) is an AI agent implementation using ADK and LiteLLM for flexible LLM integration.
Google ADK for Teaching Low Resource Languages by AI GDE Kshitiz Rimal (Nepal) describes how to use ADK and Gemini to build a personalized Nepali language tutor for English speakers.
[86+ 👏] Create Your First AI Agent ADK — Google Agent Development Kit by AI GDE Lesly Zerna (Bolivia) guides you on how to get started building your first agent using ADK.https://medium.com/media/de0f6426876b3b23b9d2746fb8bfa0b0/href
Tutorial on Creating a Drive Thru Application Using the Agent Development Kit (blog post) by AI GDE Ibnu S Wardy (Indonesia) is a tutorial to show you how to build a simple drive-thru app using ADK. From initial ADK configuration to mockups of key features like order input, order tracking, and checkout, you’ll learn the basic concepts of agent-based development and how ADK can accelerate your AI agent development process.
A Developer’s Guide to Streamlined Data Reporting (code) by AI GDE Bukempas (Türkiye) is an agent that makes the report of a dataset in BigQuery according to the request. He used ADK to build the app, Gemini 2.5 to improve codes, and Jules for Readme on Github.
https://medium.com/media/aef2696c0860a502e9717cdaac3dc748/href
Google ADK-powered solution for transnational environmental coordination (video | code | demo) by Zaynul Abedin Miah (Bangladesh) is a solution to combat cross-border air pollution, focusing on the Indo-Gangetic Plain. This agent system provides real-time monitoring, analysis, and alerting and automated policy recommendations.
Gemini
Vibe Coding (slides) by AI GDE Martin Andrews (Singapore) was a workshop introducing the Vibe Coding technique. It guided participants through a two-stage design and build process using a pre-built hinting file for the LLM, and allowed participants to upload their apps to a public site for sharing. Martin demonstrated how to use Gemini 2.5 Flash for product requirement documents and initial design and Gemini 2.5 Pro for code generation.
Several GDEs shared Gemini CLI introduction and tutorials with practical examples:
- Google’s Free Open-Source Coding Agent by AI GDE Muhammad Farooq (US)
- A Guide With Practical Examples by AI GDE Aashi Dutt (India)
- Let’s try writing Google ADK AI Agent code with Gemini CLI! by AI GDE Liu Yu-Wei (Taiwan)
- Use Vibe Coding to Build Your Own App: My Gemini CLI Fitness Log Practice by AI GDE Evan Lin (Taiwan)
Memory AI Chatbot Implementation with Gemini and LangChain (Colab Notebook) by AI GDE Rendy B. Junior (Indonesia) explores the types of memory within LangChain and discusses advantages and disadvantages of each option from keeping everything to fetching only the last 5 conversations, for example.
SmartStock Assistant by AI GDE Charmi Chokshi (Canada) is an agentic AI-powered web application that analyzes stock performance using real-time data, news sentiment (powered by Gemini 2.5 Pro), and optional chart analysis. It generates downloadable PDF reports and is deployable on Cloud Run using Docker.
Using Gemini Pro to Analyze Option Data by AI GDE Yucheng Wang (China) is a notebook to analyze and visualize options trading data using Gemini 2.5 Pro via the Vertex AI SDK in a Colab Enterprise environment. It showcases how to generate insightful commentary using Gemini’s capabilities and integrates with Pandas and Matplotlib for data manipulation and plotting.
Cardex by AI GDE Xavier Portilla Edo (Spain) is a web application for managing your Pokémon card collection with AI-powered card scanning and identification (Gemini Vision).
How do we see the world: Equation, Model, Quantum (slides) by AI GDE Svetlana Meissner (Germany) explored the relationship between equations and models in ML. She discussed building models, the measurement tools used, and the rapid growth in the number of available models covering products in the Gemini ecosystem.
Building a Burmese Language Chatbot for Local Businesses with Gemini API by AI GDE Aye Hninn Khine (Thailand) demonstrates how to build a basic Burmese language chatbot using the Gemini API to provide customer support for local businesses, like a restaurant in Yangon.
https://medium.com/media/4218746c66c49aac8a24de843af351bd/href
Vibe coding with Gemini using Google AI Studio by AI GDE Will Huang (Taiwan) is a video introducing vibe coding even for non-programmers. He emphasizes on security risks and shares how to use source code downloading and deploying safely.
Gemini’s URL Context Tool and LangChainJS by AI GDE Allen Firstenberg (US) explores how to use URL Context Tool with LangChainJS to improve the accuracy of LLM responses by grounding them in specific web pages.
https://medium.com/media/cb6fced821e0de19a033d1cc24f3f00c/href
You’re Behind If You’re Not Using These 4 New Google AI Tools by AI GDE Daniel Gwerzman (UK) explores 4 new tools from Google I/O 2025 that allow anyone, even non-developers, to build app prototypes. Starting with Gemini (the chat) and moving to AI Studio, Firebase Studio, and Jules.
#Gemini Diffusion Gemini Diffusion Is CRAZY Fast — But Not What You Think by AI GDE Muhammad Farooq (US) breaks down how it works, why it’s fast, and what it means for the future of LLMs.
#AlphaEvolve An EVOLVING AI? Meet AlphaEvolve | One step closer to AGI by AI GDE Carlos Alarcon (Colombia) explores how AlphaEvolve is solving complex math problems, evolving code-writing, and accelerating training of AI models.
#Stitch Vibe Designing with Stitch by Angular GDE Connie Leung (Hong Kong) is a video introducing Stitch and demonstrating how to create 4 UIs for an application in an hour.
#Stitch The END of designers? Stitch AI: Design UI/UX interfaces for FREE with a Prompt! by AI GDE Carlos Alarcón (Colombia) is a tutorial video showing how to create UI/UX designs from scratch using natural language.
AI Research
#Gemini GPU Kernel Scientist: An LLM-Driven Framework for Iterative Kernel Optimization by AI GDE Martin Andrews (Singapore) introduces an automated methodology powered by LLMs to address the complex challenge of optimizing GPU kernels for high performance. Gemini 2.5 Pro was used as the LLM Kernel writer and Gemini 2.5 Flash chose where to focus, and what experiments to perform at each step . The paper was accepted as a workshop paper at ICML 2025. Martin gave talks about his research and recent research topics under the following titles — Evolving GPU Kernels (covering AlphaEvolve) and Self-Improving Agents.
#Gemma Collaboration is all you need: LLM Assisted Safe Code Translation by AI GDE Rabimba Karanjai (US) is a paper introducing UniTranslator, based on the Gemma family of models, a framework for code generation as a collaborative endeavor among multiple LLMs.
#Gemma Understanding Gemma 3n: How MatFormer Gives You Many Models in One by AI GDE Rishiraj Acharya (India) provides an explanation of Gemma 3n and its underlying Matryoshka Transformer (MatFormer) architecture. It details how MatFormer allows for training one large model and deriving a family of smaller, high-performing models.
Gemma
Extracting ICD-10 Codes from Clinical Notes Using MedGemma
Extracting ICD-10 Codes from Clinical Notes Using MedGemma (Kaggle notebook) by AI GDE Gabriel Preda (Romania) shows how to use MedGemma to extract from the clinical notes the relevant ICD-10 codes. ICD-10 is a global standard diagnostic tool for epidemiology, health management, and clinical purposes, developed and maintained by WHO. Gabriel also shared MedSight, a project using MedGemma to interpret a wide range of medical images — including X-rays, MRIs, CTs, skin scans, and histopathology slides.
MedAssist: Automated Diagnostic Reports for Medical Images Using MedGemma & Fine-Tuning (code) by AI GDE Roushanak R (UK) analyzes medical scans (MRI, X-rays, CT scans) to produce detailed, expert-level diagnostic reports in seconds. Powered by MedGemma, it assists healthcare professionals by speeding up diagnosis and ensuring no subtle findings are overlooked.
[117+ 👏] TxGemma: LLM for Drug Development by AI GDE Jéssica Costa (Brazil) explains the objectives of the TxGemma model for the biological area based on the Deepmind paper. It provides a small tutorial on how to deploy the model in Vertex AI.I built the ultimate tool to know what to invest in — and what to avoid. And it’s totally Open-Source! (code) by AI GDE Eric Risco (Andorra) introduces “FinanceAI Research Platform” which he built to turn your laptop into a Bloomberg-grade research desk. Powered by DeepSeek R1, Gemma 3, LangGraph and a modern TypeScript / Next.js stack, it delivers institution-level reports in minutes.
Gemma 3n models made possible the full AI stack entirely on mobile! (code | demo video) by AI GDE Georgios Soloupis (Greece) explores how to run the full AI stack entirely on a mobile device, covering speech-to-text, function calling, VLM inference, and text-to-speech in a single android application.
Cactus: Framework for running models like Gemma on Android devices by AI GDE Henry Ndubuaku (UK) is a framework for running AI models on mobile devices, with simple and consistent APIs across C/C++, Dart/Flutter and Ts/React-Native. The demo app allows users to try variants of Gemma on their android devices.
Building an On-Device Gemma 3 Chat App with Flutter by AI GDE Sungmin Han (Korea) explores the implementation of a Flutter chat application running a 1B parameter Gemma 3 model entirely on an iPhone. It highlights the challenges and solutions involved in achieving on-device AI, including the necessity of using Flutter’s master channel, enabling the native-assets feature, and manually linking MediaPipe dependencies via a custom Podfile.
Hands-On with Gemma 3: Build Intelligent Agents from Scratch by TFUG Islamabad explored Gemma 3, its core architecture, and how to build an intelligent agent from scratch and practical deployment strategies using Colab, Hugging Face, and local environments.
Firebase Studio
From Idea to App in Under an Hour with Firebase Studio by AI GDE Siddhant Agarwal (India) is a quick walkthrough of building a QR code generator app (QRGen) using Firebase Studio. It details the experience from initial idea to deployment, highlighting the ease of use, Firebase integration, and developer-friendliness of Firebase Studio.
Firebase Studio Open Source Project Implementation — Dockerimon: Docker Monitoring Platform by AI GDE Liu Yu-Wei (Taiwan) introduces a Docker monitoring platform built using Firebase Studio. Dockerimon aims to simplify Docker management for developers by providing a local dashboard that integrates common operations, status monitoring, and log tracking. It is designed to replace common Docker commands, reduce command-line usage, and enhance visualization for DevOps, backend, and AI engineers.
Prototype, Build, and Run Full-Stack AI Apps with Firebase Studio (codelab) by AI GDE Vrijraj Singh (India) delivered a session & workshop on how Firebase Studio enables developers to quickly prototype, build, and deploy full-stack Next.js applications directly from the browser. Vrijraj highlighted the App Prototyping agent and Gemini AI integration, which allow developers to generate functional apps from natural language, images, or sketches. He also shared a collection of codelabs focused on Gemini with JavaScript.
GlamMEai by AI GDE Pankaj Rai (India) is a web app, powered by Gemini and Firebase Studio, to provide instant feedback on clothing clothing combinations. It also enables virtual try-on and transform your photos. Pankaj’s another app, NutriMonitor+ analyzes an image of food or ingredient list and provides information about the food you eat. Given your medical conditions, it also offers personalized recommendations helping you determine what to eat.
Create MCP Server and MCP Client in #Firebase studio by AI GDE Cyrus Wong (Hong Kong) details the process of setting up an MCP server and client using the Python SDK for Firebase. His another video How to use Agent Starter Pack Multimodal Live Agent plus MCP Server in Firebase Studio? focuses on integrating an MCP server with a Robot Agent in Firebase Studio.
The AI Revolution: Navigating the Future of Software and Engineering by TFUG Islamabad shared insights into building an real-world app with Gemini API, leveraging Firebase Studio for AI-powered workflows, and shifting from traditional coding to AI system-level thinking.
JAX
Deep Thinking: Keras, JAX and the Rise of AI Agents (slides | demo) by AI GDE Ahirton Lopes (Brazil) covered the Google I/O announcements. With a deep dive into the AI-related updates, he explored the critical role of Keras (including Keras Recommenders, KerasNLP, KerasCV) and JAX in building scalable, high-performance models across devices.
Ankh3: Multi-Task Pretraining with Sequence Denoising and Completion Enhances Protein Representations by AI GDE Ahmed Elnaggar (India) is an Protein Language Models (PLM) with TPUs. Using TPUs, JAX, and T5x, Ankh3 leverages multi-task pretraining combining sequence denoising and sequence completion to learn richer protein representations.
Kanana Language Model Training (slides) by AI GDE Minho Ryu (Korea) shared his insights on training an LLM with JAX on TPUs, primarily focusing on the MaxText framework. He specifically highlighted the shift from Mesh Transformer JAX to MaxText.
NanoLLM (Colab notebook) by AI GDE Saurav Maheshkar (UK) is a TPU-compatible, nano-scale LLM training codebase built with JAX/Flax. It offers quickstarts on Colab (TPUv2–8), sharding support, and checkpoint management with Weights & Biases and Hugging Face.
Keras
[153+ 👏] Make a Recommender System from Your Dataset using Keras by AI GDE Joan Santoso (Indonesia) is a tutorial explaining how to build a game recommendation system using a Steam video game dataset, leveraging Keras and the keras-rs library. It covers data preprocessing, building a retrieval-based recommender model, and generating recommendations.Colab
https://medium.com/media/c7a4cc08497bed7b0b2a8a5e7e5b69a2/href
Getting started with Google’s Data Science Agent in Colab by AI GDE Carla Vieira (Brazil) introduces how to use Data Science Agent in Colab to streamline data science workflows and tedious tasks such as suggesting optimized code, helping troubleshoot errors, and more.
Kaggle
The Evolving Landscape of Kaggle Competitions by AI GDE Gabriel Preda (Romania) is an article tracing Kaggle’s journey and explores the evolution of Kaggle competitions from simple submission-based tasks to code competitions that encourages open-ended exploration and collaborative projects.
Cloud & VertexAI
How AI Agents Are Revolutionizing Businesses and Public Administration (slides) by AI GDE Svetlana Meissner (Germany) outlines a vision of the future in which AI agents act as digital colleagues in the administration. The focus is on the collaborative strength of AI, which not only increases efficiency but also enhances citizen-friendliness and transparency. The talk included the concept of agents including A2A and MCP, and a demo of Gemini in VertexAI. She also created Potcast on the topic in English | German.
https://medium.com/media/40825e63313372149278aedb243b74bf/href
Google Cloud Conversational AI with DialogFlow CX by AI GDE Rodgers Ndocha (Kenya) explores how to build conversational AI agents using Conversational Agents (formerly Dialogflow CX). A demonstration includes creating an agent that can answer questions based on the provided PDF documents.
AI Flutter Club | Week 4: Integrating AI into Flutter Apps (video) by GDG Ahlen was a community-driven event series designed for developers keen to explore the intersection of Flutter and Generative AI. The session covered how to connect Flutter with Vertex AI and how to use AI APIs in mobile applications.
AI Agents & The Future of Coding: A Conversation with a Googler by AI GDE Muhammad Farooq (US) is an interview video with Karl Weinmeister (Head of Cloud Product DevRel at Google) discussing the role of AI agents in coding assistance and the evolving role of developers. They delved into ADK and talked about the practical applications and future of AI agents in helping developers build scalable solutions.
[Jun 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.
