[Nov 2025] AI Community — Activity Highlights and Achievements

[Nov 2025] AI Community — Activity Highlights and Achievements

We love sharing the accomplishments of the Google AI communities over the month. We appreciate all the hard work and dedication of our community members. Without further ado, here are the key highlights!

Antigravity • G3mini 🥰

[🩷284+] Building a Copado MCP Server with AntiGravity in 30 Minutes (Post + Video) by AI GDE Gaurav Kheterpal (India) shares building a simple Copado MCP server using Antigravity. It took 30 mins to get working end-to-end for a basic use case and showed Antigravity’s ability to solve problems faced during the process. He also shared his experience in a blog post (repository) with details.

[👏107+] Experience AI-powered personal website creation directly using the Google Antigravity IDE tool by AI GDE Yu-Wei Liu (Taiwan) explored Antigravity, finding it to be a full-fledged development platform built around AI agents.

The Era of Action with Gemini 3 Pro & Google Antigravity

[👏67+] The Era of Action Model with Gemini 3 Pro & Google Antigravity by AI GDE Thomas Chong (Canada) highlights the Gemini 3 Pro’s Sparse MoE architecture, performance benchmarks, and new API features (like thinking_config and media_resolution). He also introduced Antigravity as an agent-first IDE and discussed safety considerations.

[👏32+] Running Antigravity on ChromeOS / ChromeOS Flex by AI GDE William McLean (US) provides a solution for developing apps with the latest agents on Chromebooks and offers an option for running Antigravity on WSL for Windows users.

Google Antigravity Tutorial: Build a Finance Risk Dashboard by AI GDE Aashi Dutt (India) shared how Antigravity converts prompts into apps, and built a finance dashboard using Gemini 3.

Enterprise level Agentic dev workflow with Gemini CLI, ADK, and Antigravity (slides | blog post) by AI GDE Jimmy Liao (Taiwan) implemented an ADK-compliant AI agent using a complete, hands-on guide, beginning development with the Gemini CLI.

Developing a Variational Autoencoder in JAX using Antigravity

Developing a Variational Autoencoder in JAX using Antigravity by AI GDE Rubens Zimbres (Brazil) shows how to build a VAE in JAX using Antigravity, make inference, and leverage the efficiency/speed of JAX.

Build Epic with Google Antigravity (video) by AI Hajipur (India) introduces Antigravity with demonstration of autonomous code generation, bug fixing, and real-time app previewing.

[👏135+] Gemini 3.0 Pro Model Introduction — The Next Generation of Google’s Closed-Source Model Officially Launched by AI GDE Yu-Wei Liu (Taiwan) introduces the new model and its performance with outlining how the launch of Gemini 3 Pro represents a strategic evolution toward advanced multimodal reasoning and agent-driven capabilities.

Nano Banana Pro🍌

Visualizing Research: How I Use Gemini 3.0 to Turn Papers into Comics by AI GDE Grigory Sapunov (UK) details a two-step workflow for generating graphic novels from research papers using Gemini 3 Pro for text creation and Nano Banana Pro for images.

https://medium.com/media/879620013acb01026018ee82cd5e242a/href

[🩷157+] Nano Banana Pro + Google Street View API (live demo) by AI GDE Radostin Cholakov (US) demonstrates how you can create hyper-realistic photos at any location using Nano Banana Pro and Google Maps Street View API.

Nano Banana Pro on StructBench by AI GDE Sayak Paul (India) evaluated Nano Banana Pro on its ability to generate and edit non-natural images and shared the findings from this evaluation.

I Built AI That Sees Through Streets (Multi-Agent Demo) by AI GDE Noble Ackerson (US) is a teaser video of his multi-agent project. It uses three agents (seeing, reasoning, and generating) powered by Gemini 3 & Nano Banana Pro. The agents communicate via the A2A protocol and are deployed to Vertex AI Agent Engine.

Virtually Try Clothes on with Gemini Nano Banana Via RCS by AI GDE Amanda Cavallaro (UK) demonstrates how to use Nano Banana (Gemini 2.5 Flash Image) to generate and edit images, allowing users to virtually try on clothes by combining a selfie with a picture of clothing. The app allows users to virtually try on clothes by sending images to an RCS Business Messaging agent. The server uses Gemini to composite the garment onto the person in the selfie and sends the resulting image back to the user.

AI GDEs 💗 YouTube

https://medium.com/media/301ddf5532120b35365e47dd2899f1da/href

AI GDE Muhammad Farooq (US) | Go to the channel

• [👀14.1k+] Gemini 3 Pro: It was worth the wait!!!
• [👀26.4k+] Google Antigravity — Did Google Just Killed Cursor?
• [👀7.4k+] Google’s nano banana pro just killed Photoshop…

AI GDE Sam Witteveen (Singapore) | Go to the channel

• [👀98.4k+] Gemini 3 Pro — The Model You’ve Been Waiting For
• [👀35.8k+] Antigravity Google’s Cursor Killer
• [👀32.8k+] Nano Banana Pro has arrived!!

AI GDE Carlos Alarcón (Colombia) | Go to the channel

• [👀2.1k+] New features in the Gemini 3 API: Step-by-step tutorial
• [👀9.5k+] Antigravity Tutorial: Google’s New Agent IDE — FREE!

https://medium.com/media/1d05fda5d50341f351339c31816f40e6/href

JAX

[PR] Optimized MaxText LLM Data Pipeline with Efficient Sequence Chunking by AI GDE Minho Ryu (Korea) introduces “sequence chunking” using Grain’s FlatMapTransform. It prevents data loss from long document truncation during pre-training and ensures that 100% of tokenized data is utilized for more efficient training and higher-quality, unbiased LLMs.

Tools For Fine-Tuning LLMs with Tunix by AI GDE Henry Ndubuaku (UK) added Tunix from the JAX ecosystem as the official fine-tuning framework for the Cactus inference engine.

JAX DevLab 2025 Marin and Levanter by AI GDE David Hall (US) delivered a quick overview of Marin and Levanter (an open lab for open foundation model development and the training framework based JAX and Equinox), detailing the project goals, framework, and latest model.

Contributing to the JAX Ecosystem (slides) at JAX & OpenXLA DevLab by AI GDE Minho Ryu (Korea) shared his journey as an AI GDE & open-source contributor specialized in JAX. He highlighted his contributions to MaxTex (PRs: 1, 2, 3), Grain, and ArrayRecord and encouraged the audience to participate in the JAX community.

ADK • A2A

[Codelab] Building a Hiring Agent with ADK and Gemini 3 by AI GDE Vrijraj Singh (India) and AI GDE Jay Thakkar (India) guides how to make an automated workflow performing an entire hiring process from parsing resumes, extracting experience/skills, ranking candidates to generating hiring recommendations, and etc.

Introduction to Test Features for Evaluating the Deployability of AI Agents by AI GDE Yu-Wei Liu (Taiwan) introduces ADK’s AI Agent evaluation, which focuses on testing, systematically measuring, comparing, and optimizing AI agent performance for deployment readiness. It covers design principles and practical functions of ADK evaluation, emphasizing its role in maintaining quality control and development confidence in probabilistic environments.

[👏56+] Let’s use the Google ADK to develop AI agents using the YAML format! by AI GDE Yu-Wei Liu (Taiwan) introduces how to build AI Agents using ADK by defining the agent’s configuration in a declarative YAML format instead of traditional code.

A2A “Polyglot” Cross Language Development by AI GDE William McLean (US) demonstrates the use of the A2A protocol for cross-language (Python, Go, JavaScript) agent development, highlighting the flexibility of this language-neutral protocol. Following it, he evaluates the performance of the agents in Benchmarking Agents with A2A, MCP, and Gemini CLI.

ADK’s Visual Agent Builder

[👏181+] Building AI Agents Visually with Google ADK Visual Agent Builder by AI GDE Thomas Chong (Canada) explores the ADK’s Visual Agent Builder, a browser-based interface that simplifies multi-agent system architecture through a visual workflow designer, configuration editor, and AI assistant for rapid prototyping and collaboration. He also shared [👀3.4k+] a tutorial video and demo codes.

[👏66+] Hacking ADK’s Importer: How We Slashed 24-Second Cold Start in Half by AI GDE Marton Kodok (Romania) explores reducing cold start times for an AI agent built with ADK framework deployed on Cloud Run. It details an experimental workaround to optimize the ADK importer and significantly reduce the unacceptable initial 24-second cold start time.

Google ADK Visual Agent Builder Tutorial With Demo Project: Travel Planner by AI GDE Aashi Dutt (India) shows how to build a Gemini-powered travel planner using the ADK’s Visual Agent Builder to design sub-agents, wire tools, and ship an itinerary workflow in minutes.

Gemini

Build a Smart File Assistant LINE Bot with Python + Gemini File Search: Let AI Read Your Files for You by AI GDE Evan Lin (Taiwan) describes how to build a LINE bot using the Gemini File Search API to answer user questions about uploaded documents (PDF, DOCX, TXT, images). It details the implementation of key features like multi-dialogue isolation, file management, and intelligent error handling.

[👀10.8k+] RAG with Gemini: Goodbye to Vector Databases! (Colab notebook) by AI GDE Carlos Alarcon (Colombia) is a tutorial video guiding you how to create an advanced RAG system without the complexity of managing vector databases, manual chunking, or embedding. He demonstrates how the Gemini File Search API automates the entire process.

[👀8.6k+] Google File Search API — Managed RAG for Everyone by AI GDE Muhammad Farooq (US) demonstrates a production-ready app with citations, hybrid search, progressive indexing, Clerk auth + Firebase, and breakdown pricing.

https://medium.com/media/d85fa3378f2e0842c8fb2679868e7b4b/href

[👀44.9k+] Gemini RAG — File Search Tool by AI GDE Sam Witteveen (Singapore) goes through the Gemini File Search API, a simple RAG system built into the API, and demos of simple & advanced version (Colab notebook 1- simple| 2- advanced)

Gemini CLI

[Codelab] Gemini CLI: The All Rounder Developer friend in the terminal by AI GDE Jay Thakkar (India) walk you through real developer workflows: automating code reviews, integrating with CI/CD pipelines, generating/running complex shell commands, creating documentation and tests, and etc.

LeapCode Multi-AI Coding CLI Wrapper (Now support Gemini-CLI) by AI GDE Jimmy Liao (Taiwan) currently supports Gemini CLI and provides a unified interface for various AI tools, I/O interception, and etc.

Gemma

Reason with Your Images at Home: Use Gemma 3 as Visual Thinking Model by AI GDE Ertuğrul Demir (Türkiye) explores the application of reasoning to VLMs using RL and optimization tricks. The post details a three-stage structured generation framework (detection-thinking-answer) to prevent overlooking visual details. It also covers GRPO’s role in teaching structure without hard labels, improvements with GSPO and Dr. GRPO, LoRA rank-1 sufficiency, and training VLMs at home using Unsloth, vLLM, and TRL.

Run EmbeddingGemma with ONNX on mobile by AI GDE Georgios Soloupis (Greece) showcases how to use EmbeddingGemma in the .onnx format inside a mobile to power an AI system. This app, which uses RAG, provides highly accurate, relevant, and context-aware answers to user queries.

Keras

[PR] AI GDE Hongyu Chiu (Taiwan) added DINOV3 to KerasHub with assistance from the Gemini CLI. Additionally, he shared detailed Gemini CLI instructions for implementing DINOV3 here: Add the workflow for co-working with the Gemini CLI.

Colab

Demystifying Text-to-Speech and Voice Cloning for Content Creation (Colab notebook | repository) by AI GDE Muhammad Ghifary (Indonesia) explores how modern TTS and voice cloning can make content creation easier and personalized. The article covers the evolution of TTS from early mechanical speech machines, rule-based systems and concatenative synthesis, to today’s neural pipelines, and introduces voice cloning as the next frontier: preserving “what is said” while mimicking “how it’s said”. It then presents a prototype using Chirp 3 for TTS and Gradio for creating the UI.

Colab AI First Google Data Science Agent (Colab notebook | slides) by AI GDE Burak PASTIRMACI (Türkiye) shared how Colab’s AI-powered Data Science Agent enables using natural language for faster data cleaning, visualizing insights, and building models.

Colab VS Code Extension Sprint projects

[👏201+] Google Colab VSCode Extension: Goodbye to Context Switching by AI GDE Ifeanyi Idiaye (Nigeria) is a quick tutorial for Google Colab Extension guiding you how to install and integrate with Gemini CLI.

[👏95+] Run Google Colab Inside VS Code + Analyze Video with Qwen3-VL & FiftyOne by AI GDE Adonai Vera (Colombia) walks through how to use the new Colab extension for Visual Studio Code to connect to a Colab GPU/TPU runtime directly, and then utilizes this setup to run a video-analysis workflow with Qwen3‑VL and FiftyOne.

[👏67+] Install and launch the Google Colab Extension in VS Code to run the Google Gemma 3 VLLM model service (Colab notebook) by AI GDE Yu-Wei Liu (Taiwan) explains how to use the Google Colab extension in VS Code to run the vLLM-based service of Gemma 3. It details setup steps and benefits like a unified workflow and GPU/TPU access.

Fine Tuning VaultGemma with Differential Privacy using a Colab Runtime in VSCode by AI GDE Rubens Zimbres (Brazil) is a full guide on building private AI to address LLMs’ tendency to leak sensitive data. It details how to use VaultGemma, add a layer of privacy, and run the entire workflow in VS Code connected to a free Colab T4 GPU.

[👏52+] Connect to Google Colab Server in VS Code IDE with Colab Extension by AI GDE Kuan Hoong Poo (Malaysia) outlines how the new Colab extension for VS Code allows connecting to Colab servers within the IDE, combining a full-featured editor with cloud-based compute (GPU/TPU access) for AI/ML workflows.

Others

[🩷284+] Building a Flutter App in 10 Minutes with Google Stitch by AI GDE Didier Girard (France) shares his experience of creating a functional Flutter app for tracking blood pressure in under 10 minutes using Deepmind’s Stitch for rapid UI prototyping.

https://medium.com/media/9aece96f7e73d2e30f8e3f86d7ef6f72/href

The ML Math Club by GDG On Campus ISAMM (Tunisia) completed five sessions throughout the year helping students master math concepts for AI in both Tunisian Dialect and English.

AI Research Paper Reading Club 2025 by ML Nashik (India) covered foundational/classic papers and current trends/apps, alongside training in professional writing aimed at publishing in conferences.

---

[Nov 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.