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!
Product highlights
Gemini 3.1 Pro
https://medium.com/media/f2c4d7b237ceba04efb1fef29c870812/href
[👀34.5k+] Gemini 3.1 Pro: The model no one expected by AI GDE Muhammad Farooq (US) breaks down why this release is a huge leap forward for AI with the benchmark scores. In addition, he discusses [👀45.3k+] The 100x AI Breakthrough No One is Talking About, introducing Aletheia (the new math research agent from Google DeepMind) and explaining why “thinking time” is the new scaling law and its impact on the future of AI development.Tutorial: Get started with Google Gemini 3.1 Pro by AI GDE Tarun Jain (India) explores what has been updated in the new model including thinking model, custom tool usage, multimodality, etc.
Gemini
Gemini 3 Agentic Vision: Moving Beyond Simple Image Prompting (Colab notebook) by AI GDE Connie Leung (Hong Kong) details how to build a “Digital Docent” using Python and Gemini SDK to enhance visual accessibility for users with low vision. The model employs a Think-Act-Observe loop, autonomously writing and executing Python code to crop and enhance specific areas based on user prompts.
[👀18.9k+] Gemini’s Native Web Scraper: 100% “Free” & Multimodal by AI GDE Muhammad Farooq (US) explores the Gemini URL context tool for scraping data from the web, including images and pdfs. This video demonstrates how to set up and use the Gemini API with practical examples and functionality enhancements.Gemini CLI
Agentic AI
WebMCP
WebMCP: Chrome Just Turned Every Website Into an API for AI Agents by AI GDE Rohit Ghumare (UK) introduces WebMCP, a new API that allows websites to register structured tools in the browser, enabling AI agents to interact without screenshots or scraping. This approach helps reduce token usage and improves accuracy while offering JavaScript and HTML-based tool registration.
ADK
Skills, Not Vibes: Teaching AI Agents to Write Clean Code (repository) by AI GDE Ertuğrul Demir (Turkey) explores the importance of teaching AI agents to write clean code by applying the Clean Code principles through instruction files called Skills. It highlights the rise in code complexity due to AI adoption and emphasizes using control systems to maintain quality.
From Single Agent to Production-Ready Multi Agent Architecture with ADK, MCP, A2A, and AG-UI by AI GDE Henry Ruiz (US) explains how to evolve a single-agent AI application (continuous from his previous article) into a scalable multi-agent system using ADK, MCP, A2A, and AG-UI protocols. It demonstrates how decomposing logic into specialized agents improves modularity, scalability, and maintainability for complex tasks.
[Tutorials] Developer Knowledge API
Google Developer Knowledge API and MCP Server: Install the official knowledge base for your AI assistant by AI GDE Evan Lin (Taiwan) discusses the API and the MCP server, which provide AI assistants with real-time access to machine-readable official documentation. It explains how these tools improve AI accuracy and includes a guide for configuring Gemini CLI with the Knowledge MCP Server.
Gemma
An offline translation web service built on TranslateGemma (Colab notebook | repository) by AI GDE Yu-wei Liu (Taiwan) introduces a new translation model, which supports 55 languages and image input. It also covers a personal project to create an offline, secure translation webpage inspired by the model, offering practical examples to try out.
[👩🏻💻Codelab] From Tokens to Tools: Exploring Gemma 3 by AI GDE Krupa Galiya (India) is a codelab walking you through using Gemma 3 with Hugging Face. It covers practical experimentation including environment setup, model loading, tokenization, and text generation. It also demonstrates structured outputs and tool (function) calling, integrating the model with real Python functions for tasks like weather lookup, currency conversion, and etc.The “Soul of Taiwan” Defending Digital Frontiers: A Deep Dive into the T1 Series — A Google Gemma 3 Localized Model for Taiwan by AI GDE Jerry Wu (Taiwan) introduces Twinkle AI (based on Gemma 3–4B-T1-it) and emphasizes the importance of Sovereign AI. The model excels in deep localization, robust AI agent potential through optimized function calling, and lightweight deployment for data security, outperforming the original Gemma 3 in Taiwan-centric tasks.
Pre-placement Hands-on Workshop (slides) by AI GDE Tarun Jain (India) was a 2-day event that covered: use of system prompt and different prompt techniques; inferencing of open source LLM (Qwen and Gemma 3) using HuggingFace and vLLM; and practical guides to implement RAG systems (Colab notebook).
https://medium.com/media/d10e9d2804eb0ab3bb8ce48f0515c372/href
Using Android Studio and VSCode to Code with Offline AI Models (video) by TFUG Islamabad (Pakistan) and AI GDE Georgios Soloupis demonstrates how to build a secure, fully offline coding workflow by integrating Gemma into Android Studio and VSCode for total independence from cloud tools.
Cloud
From Proof of Concept to Production: Building an Enterprise-Grade Platform for AI Systems (repository) by AI GDE Rubens Zimbres (Brazil) presents a comprehensive reference architecture for deploying multi-agents AI systems on GCP, designed with the explicit goal of allowing developers to plug any AI agent system into a robust infrastructure.
How We Halved Latency in PHP with BigQuery Short Query Mode by AI GDE Marton Kodok (Romania) demonstrates how to reduce query latency by over 50% using SQM. It explains how bypassing the traditional Job creation and polling lifecycle for small, frequent queries can significantly reduce orchestration overhead.
JAX & TPU
Why JAX? The NumPy You Know, But Faster by AI GDE Wesley Kambale (Uganda) explains how JAX uses XLA to accelerate mathematical operations. It serves as a beginner-friendly introduction to the framework’s performance capabilities.
Running Native PyTorch on TPUs with Zero Code Changes by AI GDE Rishiraj Acharya (India) introduces TorchTPU to enable running unmodified PyTorch models on TPUs with high performance and a focus on native developer experience.
JAX 0.9 Release Notes Summary by AI GDE Sho Tanaka (Japan) summarizes JAX 0.9.0 updates including Effort-Based Versioning (EffVer), new features like jax.thread_guard, and explicit sharding support in jax.export. It also discusses the migration from pmap to shard_map.
Automatic vectorization with vmap and gradients with grad in Jax by AI GDE Wesley Kambale (Uganda) covers the core JAX transforms of jit for speed, vmap for batching, and grad for gradients to train neural networks from scratch.
dl-jax101 by AI GDE Muhammad Ghifary (Indonesia) is a learning resource (written in Indonesian) for understanding and implementing deep learning concepts using the JAX framework. It focuses on the optimization and high performance offered by the JAX ecosystem (including Flax and Optax).
Nano-MoE-JAX by AI GDE Kartikey Rawat (India) is a lightweight, educational Mixture-of-Experts (MoE) GPT-style language model (inspired by nanoGPT) built from scratch in JAX/Flax.
Colab
Free Academic Paper Translation with TranslateGemma (Colab notebook) by AI GDE Jimmy Liao (Taiwan) introduces TranslateGemma, a tool for translating arXiv papers into bilingual HTML using Colab’s free T4 GPU. The article provides a step-by-step guide on setup, configuration, and result viewing while discussing translation quality and performance.
ODML
#JAX [👏252+] Build and Deploying on Device ML Model by AI GDE Joan Santoso (Indonesia) explores the shift to ODML using JAX, Flax NNX, and AI Edge LiteRT to build a digit-recognition Flutter app. It demonstrates a practical workflow for model conversion and envisions a future of specialized, resource-efficient on-device AI agents. He also gave a talk on the topic, Best Practice to Deploying Model to On Device ML (slides), with a live demo component for the Flutter integration.
[👏123+] On-Device Function Calling with FunctionGemma by AI GDE Sasha Denisov (Germany) introduces FunctionGemma and compares it with other competitors in detail with guidance for mobile/web developers.Community Spotlights
Google Agentic AI Engineer | ADK, Vertex AI & Multi-Agent | Udacity
AI GDE Allen Firstenberg (US) and Noble Ackerson (US) co-launched Google Agentic AI Engineer (Udacity’s Nanodegree) with 5 other instructors. This program (55 hours across 4 courses with 4 capstone projects) covers advanced prompting, agentic workflows, and how to build RAG and multi-agent systems using Gemini, ADK, VertexAI, and GCP infrastructure.
AI Training Campaigns
https://medium.com/media/e74bff3b91885366ced8c3e4c35ac689/href
Math for ML by Tanmoy Tanoy (Bangladesh) completed 4 sessions delivering foundational mathematical pillars for ML and reaching 3500+ views.
AI Paper Reading Clubs (playlist) by TFUG Islamabad (Pakistan) covered advanced AI research, ranging from CLIP’s visual-language learning to explainable Gen AI and distributional AGI safety.
Discretization, Outliers, Feature scaling, ML Pipelines by Machine Learning, AI, Deep Learning & NLP Community — Bangladesh (Bangladesh) explored four essential data preprocessing techniques used in ML workflows, while suggesting Vertex AI as an ideal platform.
[Feb 2026] 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.
