![[july-2025]-ai-community — activity-highlights-and-achievements](https://prodsens.live/wp-content/uploads/2025/08/38409-july-2025-ai-community-activity-highlights-and-achievements-550x300.jpg)
[July 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
https://medium.com/media/cc6c9e7c6cd06517c6b74c241180b6b4/href
Run Ollama on Colab | Open source LLM use cases | Gemma-3n (Colab Notebook) by AI GDE Tarun R Jain (India) shares how to generate a structured response using agents, use T4-GPU, and run Ollama models on Colab for free. The use cases include a multilingual translation agent, an invoice processing agent for structured data extraction from images, and a business requirements document generator.
Deploying Gemma3 with Ollama on Google Cloud Run: A Step-by-Step Guide (repository) by AI GDE Jimmy Liao (Taiwan) provides a scalable and cost-effective way to run inference. It covers cloud-based and local building options, and supports different Gemma3 model variants.
https://medium.com/media/cd5ea016c3245d81386f0e7c9a376256/href
Create a Remote LLM Server Using Kaggle Notebooks and Ollama (Kaggle Notebook | video) by AI GDE Dimitre Oliveira (Brazil) shows how to transform a Kaggle notebook into a remote server for LLMs using Ollama, eliminating cost, simplifying deployment, empowering experimentation, and enabling remote access. He has also shared how to Host Your Own Ollama Models for Free with Google Colab, a few months back.
[Post][👏73+] Google Stitch: Google’s new AI experience model for UI/UX by AI GDE Yu-Wei Liu (Taiwan) [Video] [👀134k+] Opal — Google Labs Killer NEW App by AI GDE Sam Witteveen (Singapore)Agent Developer Kit (ADK)
https://medium.com/media/e10e49095b9b7641b98bee2fcb644958/href
The AI Agent Challenge: Who Built the Best Fashion AI is the first video of Google Cloud AI Agent Build-Off series. This 3-hour hackathon-style event, designed to showcase the practical applications of ADK, challenged participants to build an AI agent that bridges the “Imagination Gap” in online fashion. AI GDE Sam Witteveen (Singapore) and Jigyasa Grover (US) participated in separate teams!
Building an agentic system with ADK and Gemini to simulate a paper review committee by AI GDE Gabriel Pereira (Romania) demonstrates building an agentic workflow with ADK to emulate/automate a paper submission review process. He used Gemini and ADK to create 4 reviewer agents and a meta-reviewer agent having different personas to build a committee. He also shared a project, Local Explorer Assistant (repository) building a day trip planner with Gemini/ADK.
https://medium.com/media/3e38b299a9102f82615878adb376e25a/href
[👏90+] Introducing Agentic Component Recognition (repository | video | Colab Notebook) by AI GDE Noble Ackerson (US) explores ACR, a concept where AI agents dynamically recognize data patterns and generate contextual UI components. The post covers the evolution of Gen UI, an orchestrator-worker architecture using ADK and Gemini/Gemma. He also addresses how to improve speed and efficiency in Evolving a Generative UI Multi-Agent Architecture through evals.Agentic AbadIA series (blog post | Español version playlist) by AI GDE Juantomas Garcia (Spain) is a series of projects aimed at solving the notoriously difficult game, “La Abadía del Crimen”. It focuses on how AI can go beyond text generation to actively play and interact with the game. Juantomas introduces and integrates AI products like Gemini/Gemma and concepts like ADK, MCP, and A2A to build a multi-agent system tackling the game’s complexities.
Implementing Anthropic’s Agent Design Patterns with Google ADK (repository) by AI GDE Henry Ruiz (US) showcases six of the foundational agent design patterns from Anthropic’s article using ADK.
Neural Reels (post | video) by AI GDE Dimitre Oliveira (Brazil) is an app that automates short-form video creation using ADK and Gemini, Imagen, and Veo. He demonstrates How AI Agents Can Revolutionize Content Creation.
Two open-source healthcare projects by AI GDE Krupa Galiya (India) using Gemini, Google Maps, ADK — 1) MediMind-AI-Multimodal-Health-Intelligence analyzing text, images, and audio to deliver health insights, from rash analysis to symptom tracking in English, Hindi, and Spanish, and 2) MediMind-AI-Agent (modular framework) providing personalized medical guidance, symptom analysis, and medical information retrieval.
Built an AI with an AI using Google ADK | Shreyak Gupta posted on the topic | LinkedIn
AI-Powered AI Development with ADK Studio by AI GDE Shreyak Gupta (US) shared a video demo of the AI that builds more AI!
[Tutorial] Building Your First Agent with Google A2A/ADK by AI GDE Jimmy Liao (Taiwan) is a hands-on guide on creating a reimbursement agent from scratch. [Tutorial] Fast and secure AI agent deployment to Google Cloud with GitLab by AI GDE Regnard Raquedan (Canada) guides how to deploy an AI agent, built with ADK, to Cloud Run using GitLab’s integrations. [Tutorial] Langchain’s SQL Agent to Google’s MCP Toolbox & ADK by Cloud GDE Jacob Mammoliti (Canada) details the setup of MCP Toolbox and the integration with ADK for building an agent. [ADK Course on YouTube (6 videos)] Master Google ADK: Practical Development Course (repository) by AI GDE Carlos Alarcon (Colombia)https://medium.com/media/83306b97148927c040a8d9a93dacfe49/href
[Talk] Securing the Future: Building Safe and Trustworthy Agents with ADK (slide) by Omotayo Aina (UK) dives deep into ADK’s security features designed to build trustworthy agentic systems. Omotaya demonstrated practical implementation of authentication controls, content safety filters, callback-based guardrails, and sandboxed execution environments.Gemini
Projects
Use hallucination as feature for vibe coding (live demo) by AI GDE Chansung Park (Korea) shows how Gen models can produce diverse web design concepts in parallel, enabling fast identification of promising directions. Once strong candidates emerge, the process shifts to lightweight refinement using smaller, cost-efficient models.
https://medium.com/media/9f3d2555933d049037cdbddaf21b67a1/href
Gemini Live Avatar: Real-time 3D Avatar Interaction with Gemini Live API (repository) by AI GDE Henry Ruiz (US)
Optimized code generation for image and video generation (repository | demo) by AI GDE Sayak Paul (India)
Gapminder-Dash (repository) by AI GDE Ifeanyi Idiaye (Nigeria) is a dashboard that enables you to query the Gapminder dataset using natural language.
Elimu Research Assistant by AI GDE Victor Ashioya (Kenya) is a tool for localized educational content creation in Kenya.
Kubernetes Log Analysis by AI GDE Yu-Wei Liu (Taiwan) helps collect and analyze Kubernetes logs using natural language, providing solutions to identified problems.
Auto Diffusers Configuration (repository) by AI GDE Chansung Park (Korea) showcase how to use Gemini to automatically configure hardware resource optimization when using Hugging Face diffusers library.
Gemini CLI
Gemini CLI with MCP Server: Expanding Possibilities with Google Apps Script
[Tutorial][👏231+] Gemini CLI with MCP Server: Expanding Possibilities with Google Apps Script by Google Workspace GDE Kanshi Tanaike (Japan) provides setup instructions and examples that allow Gemini to automate complex tasks in Google Workspace, boosting productivity. [PR-merged!] Experience of submitting Gemini CLI Pull Request using Gemini CLI by AI GDE Jimmy Liao (Taiwan) [👀137k+] Introducing Gemini CLI and [👀20k+] GeminiCLI — The Deep Dive with MCPs by AI GDE Sam Witteveen (Singapore) [⭐39+] Gemini CLI project initialization setup assistant tool by AI GDE Will Huang (Taiwan)DAmon: Fine-Tuning data arrangement Tool (code) by AI GDE Yu-Wei Liu (Taiwan)
Workshop Highlight
#AI Studio Building an AI Voice Agent to Automate a Robot Cafe with Gemini Live and Angular (slide | repository) by AI GDE Gerard Sans (UK) provided a practical hands-on guide to building a voice-first AI agent using Gemini 2.0 and Gemini Live API. Participants learned how to create a voice-automated coffee shop experience that handles complex orders and cancellations, integrating real-time voice and visual data from a webcam. Gerard has also shared a tutorial on building a Gemini Live API client from an Angular project, [👏53+] Building Advanced AI Voice Assistants Using Google Gemini 2.0 and Angular.
Gemma
[Tutorial][👏77+] Fine-tuning MedGemma 4B using Google Colab and Hugging Face with mammography images (Colab Notebook 1, 2) by AI GDE Fernanda Wanderley (Brazil) [Tutorial][👏121+] From Lab to Clinic: Unlocking Gemma 3’s latest medical AI (Colab Notebook) by AI GDE Mikaeri Ohana (Brazil) — how to deploy MedGemma for medical AI app covering implementation on Vertex AI and Hugging Face Pipelines. [Tutorial] Fine-Tuning a Vision Language Model in 2025 by AI GDE Anshuman Mishra (India) — finetuning Gemma 3 using QLoRA for VQA on chart data using the Hugging Face ecosystem. [👍57+] Testing Gemma 3n like a pro (Kaggle Notebook) and medium post by AI GDE Gabriel Preda (Romania)On-device AI
https://medium.com/media/47a26597f662e740452a759f0c6ec6ed/href
[Video] Live Captions XR (repository | project website) by AI GDE Sasha Denisov (Germany) is an accessibility project for people with disabling hearing loss. With Gemma 3n on-device and flutter_gemma library he developed, this app enables local inference within a mobile app. [Video] Medical Emergency Assistant (repository | live demo) by AI GDE Eric Risco (Andorra) is an end-to-end pipeline for building and deploying medical AI models specifically for offline emergency scenarios. [Video] AI-Powered Educational App Built with Gemini & Gemma AI by AI GDE Geeta kakrani (India) is an Flutter app designed to help students in job preparation, concept learning, and interview, all through an AI-assisted approach.Responsible AI
Reversible Anonymizer: A Python Package for Text Anonymization and Secure RAG Architecture: Reversible PII Protection for LLM Systems by AI GDE Omotayo Aina (UK) explain how to use Google Cloud’s Sensitive Data Protection to anonymize text securely and protect PII data in RAG applications.
Colab
[🩷925+] Fine-tune Gemma3n on FineVideo (Colab Notebook) by AI GDE Merve Noyan (France) [🩷400+] AutoDedup for Hugging Face Datasets (Colab Notebook) by AI GDE Maxime Labonne (UK)Firebase Studio
Keras
Keras Model Training on Cloud TPU VM by AI GDE Muhammad Ghifary (Indonesia) provides practical steps for training deep learning models built with Keras and the TF backend on a Cloud TPU VM. It includes a demonstration of how to run NeRF training building on his previous article, Implementing Neural Radiance Fields (NeRF) with Keras 3 (repository).
[💾7744+] AI GDE Harshal Janjani (India) contributed HGNetV2 to KerasHub. It’s a modern CNN architecture for edge-focused image classification. In about a month since being added to KerasHub, it has been downloaded 6000+ times.JAX
Accelerating Science with JAX — Simulations, Physics, and Beyond by AI GDE Henry Ruiz (Brazil) discusses the use of JAX for accelerating scientific simulations. It explores how JAX is reshaping modern scientific computing by enabling high-performance, differentiable, and hardware-accelerated workflows across a wide range of applications.
SoupR1 by AI GDE Anique Tahir (US) is an implementation of Qwen Models in JAX nnx.
https://medium.com/media/d71a8f60cca37957460674b6b81cec98/href
Large-Scale Language Model Training with MaxTex on TPUs (slides) by AI GDE Minho Ryu (Korea)
You’ve Got Mail 📨 (and Recommendations!): Delivering Recs with Keras, JAX & KerasRS (EN-US) by AI GDE Ahirton Lopes (Brazil)
Cloud
#Imagen4 [👏73+] Building AI Fashion House: A Multi-Agent System (repository) by AI GDE Henry Ruiz (US) and Margaret Maynard-Reid (US) built a multi-agent system that transforms design ideas into visuals. The system automates the entire pipeline to produce outputs including images and videos.
#Imagen4 Dynamic Pages with Gemini-Flash and Imagen by AI GDE Pedro Lourenço (Brazil) is a FastAPI app that generates complete web pages dynamically with generated images. Only a single keyword works to create your own custom page.
[👍85+] Content creation using Veo 3 by AI GDE Roushanak R (UK)Fashion Sketch2Runway with Gemini 2.0 Flash and Veo 3 (repository | video tutorial) by Margaret Maynard-Reid (US) and AI GDE Nitin Tiwari (India) details how to transform hand-drawn fashion sketches into realistic images/videos.
[Talk] DeepSeek R1: Advanced AI Reasoning on Google Cloud Platform and Vertex AI by AI GDE Daniel Goncharov (US) showcased integration possibilities and potential benefits for AI applications.AI Research
AlphaEvolve and Darwin Gödel Machines : LLMs for Code Evolution in 2025 by AI GDE Martin Andrews (Singapore) explores how LLMs are driving self-improving code through evolutionary methods. It details AlphaEvolve and Darwin Gödel Machines, where AI generates, evaluates, and evolves its own algorithms for enhanced performance across various tasks. Plus, Martin’s paper on verifiable reasoning for a hard NLP task was accepted to ICML 2025!
[July 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.
