[Mar 2025] AI Community — Activity Highlights and Achievements

[Mar 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!

Vertex AI Sprint 2025

The Sprint is a program that encourages the community developers to create content for specific topics within a fixed duration. This Vertex AI Sprint has been successfully completed through Feb 2025 with 98 blog articles, 35 videos, 94 code samples, and 8 AI models created by the participants (AI GDEs, Cloud GDEs, and AI Community Network)! Here are some tutorials from the sprint!

• Fine-tune Gemini to Create Reasoning Agent: A tutorial on using Gemini on Vertex AI to build a reasoning engine
• Deploy Gemma on Vertex AI in 60s: A tutorial on how to deploy Gemma on Vertex AI. This complete guide follows an end-to-end example
• Do It Yourself (DIY) MLOps on Vertex AI: A tutorial is a practical guide on how to deploy custom models on Vertex AI (Colab#1, Colab#2, Github)
• Detecting Similar SQL Queries with Vertex AI and Vector Search: A tutorial helps the readers to understand how to use embeddings and vector search while leveraging Vertex AI to find semantically similar SQL queries in a fictional dataset
• Virtual Try On with Imagen 3 on Vertex AI: A project enables virtual try-on of different clothes and outfits using Gemini for zero-shot object detection and Imagen 3 for inpainting generated images (Colab)

Kaggle Community Olympiad

The Kaggle Community Olympiad 2025 campaign is running until August. Currently, 4 Kaggle Community competitions (including 1 private) are open by the following AI communities:

• Kaggle Community Olympiad — Used Cars Prices by ML Community Talks Agadir
• Kaggle Community Olympiad — Unmasking Fakes by TFUG Prayagraj
• Kaggle Community Olympiad — Risk Analysis by ML Hajipur

Key Highlights

AI Agent

https://medium.com/media/017a34d4d23cc677d2bf114785d1f76d/href

The End of Data Science?🤯: Data Science Agent with Gemini + Colab for FREE! by AI GDE Carlos Alarcon (Colombia) is a video guide on how to use Data Science Agent in Colab. He explains how to use it without programming knowledge to extract valuable insights from data and how to visualize them.

Building a Local Agentic RAG System using Gemma-3, FastEmbed and Qdrant by AI GDE Tarun R Jain (India) explores how to build a 100% local agentic RAG system using Gemma 3 and open-source libraries. The system allows you to create a knowledge base from web data and answer user queries without relying on external APIs, ensuring data privacy and flexibility.

https://medium.com/media/1eb8f06b9a0869954106e90b40cb382e/href

Building Agent Workflows with Gemini 2.5 Pro — Does It Hold Up? (Colab notebook) by AI GDE Muhammad Farooq (US) explores Gemini 2.5 Pro’s benchmarks, coding abilities, and its performance in function calling. He demonstrates building a text-to-SQL assistant, creating complex workflows like trip planning, and simulating business intelligence dashboards.

Building a Job Matching Tool with Multi-Verbal Agent Collaboration (repository) by AI GDE Henry Ruiz (US) a talk with hands-on demo exploring how multi-verbal AI agents can collaborate to develop a simple yet effective job-matching tool from scratch without relying on existing frameworks. He focused on explaining the fundamental building blocks of multi-agent applications, including natural language understanding using LLMs, task-specific reasoning, and agent coordination.

LINE Messaging API x Vertex AI Agent Builder: Revolutionizing AI Agent Creation with Ease and Speed by AI GDE Punsiri Boonyakiat (Thailand) is about how the integration of LINE Messaging API and Vertex AI Agent Builder simplifies and accelerates the development of intelligent AI agents.

Gemini 2.5 Pro Agents (with AutoGen v0.4 Studio and No Code) (code samples) by AI GDE Victor Dibia (US) shows you how to build a travel assistant agent with AutoGen Studio powered by the new Gemini 2.5 Pro models. He also compares Gemini 2.5 and Claude 3.7 on a travel planning task.

Episode 226 — Examining Google’s Perspective on Agents by AI GDE Allen Firstenberg (US) and Mark Tucker discusses Google’s AI Agents white paper, analyzing its potential and shortcomings. They explore the core components of agents — models, tools, and orchestration — and raise questions about unclear definitions and the omission of authentication considerations.

Gemini

🦊Forky: Your AI Buddy for Open Source — Making Contributions Easier! (repository | demo) by AI GDE Adonai Vera (Colombia) makes it easy to understand repositories, navigate PRs, and get Gemini-powered suggestions to improve your contributions to open source ecosystems.

🐅 Automating Jira project management with Gemini 2.0 and Crew AI (repository) by AI GDE Hugo Zanini Gomes (Brazil) demonstrates an open-source product called TIGER (Tickets Insights Generation and Efficient Reportings) that automates Jira project management using Gemini 2.0 and Crew AI. He developed it to address time-consuming project management tasks that were cutting into productivity.

Gemini 2.5 Pro Thinking — First Look by AI GDE Muhammad Farooq (US) introduces the newly released Gemini 2.5 Pro especially with benchmark results, coding and reasoning capabilities.

Let Gemini find a summarized response through Google Custom Search based on the keywords of your question by AI GDE Evan Lin (Taiwan) guides how to use function calling with your own apps/services built on GCP with Google Custom Search API. He also shared [Gemini][MCP] Use Gemini on Cline to call the MCP function, which covers MCP, its basic principles, and how to use Cline as a MCP host with Gemini.

Prompt Engineering with Gemini Flash 2.0: From Theory to Practice by AI GDE Pedro Lourenço (Brazil) walks you through 5 prompt engineering techniques using Gemini Flash 2.0 to showcase how they work and how to choose one for your purpose.

Gemini code Assisted for Developer Productivity (slides) by AI GDE Punsiri Boonyakiat (Thailand) demonstrates how Gemini Code Assist accelerates the development lifecycle and project timelines. In the demo, she presented a use case — writing a prompt for Gemini to create a website that randomly assigns participants to Hogwarts houses and that shares the results via the LINE Messaging API.

Gemma

https://medium.com/media/3fdeb025f2eb31677250de04ce489500/href

Google’s Gemma 3: The Best Open-Weight Model Yet? by AI GDE Muhammad Farooq (US) is a video introducing Gemma 3 in detail with 4 models ranging from 1B to 27B, its features and performance, and coding capabilities.

LangChain.js Support updated to 0.2.2 by AI GDE Allen Firstenberg (US) includes Gemma 3 support for AI Studio, responseModalities (tests and code review), and tests for Gemini 2.0 Flash Lite and Gemini 2.5 Pro. This package fully supports the Gemini 2.0 conversational API in JavaScript/TypeScript.

Fine-Tuning Gemma 3 1B-IT for Financial Sentiment Analysis: A Step-by-Step Guide

Fine-Tuning Gemma 3 1B-IT for Financial Sentiment Analysis: A Step-by-Step Guide by AI GDE Luca Massaron (Italy) shows how he finetuned Gemma 3 for financial sentiment analysis using the FinancialPhraseBank dataset and achieved over 85% accuracy.

Gemma 3: The Incredible AI for PC (Code Included) 🚀✨ by AI GDE Carlos Alarcon (Colombia) explores its multimodal capabilities, expanded context window, and support for over 140 languages.

Next-Gen RAG with Couchbase and Gemma 3: Building a Scalable AI-Powered Knowledge System (repository) by AI GDE Kartikey Rawat (India) explores how to build a scalable AI-powered RAG system using Couchbase for fast and efficient vector search and Gemma for generating embeddings and responses. This step-by-step guide covers setup, document storage, embedding generation, and retrieval so that you can integrate AI-driven search into applications.

Luganda Inference on Gemma 3 (repository) by AI GDE Wesley Kambale (Uganda) is an article about the implementation of an inference in Luganda on Gemma 3 for image/video tasks using Hugging Face Transformers.

RefSheet Chat: Chat with a character via reference images! (repository | demo | slides) by AI GDE Xihan Li (China) is an open source project using multimodal capabilities of Gemma 3. It understands an image of a character’s reference sheet and talks to you just like the character.

Unboxing #gemma3 using #ollama #windows11 #AI #seeing horoscopes #fortunetelling by AI GDE Cyrus Wong (Hong Kong) shows how to use Gemma 3 for it to tell your fortune with your birth date, time and place.

JAX

JAX-Dataloader by AI GDE Kartikey Rawat (India) is a data loading library for JAX applications to load various file formats including CSV, JSON, etc. His post, Building an Efficient DataLoader in JAX: From Scratch to Production walks you through why and how he built it from scratch covering from data batching to jax.jit optimizations.

Building Convolutional Neural Networks in JAX by AI GDE Wesley Kambale (Uganda) is a walkthrough of building a CNN using JAX, from data preprocessing to training and evaluation.

Keras

PR for Qwen 2.5 on KerasHub by AI GDE Anshuman Mishra (India) contributed to the open-source ecosystem by adding Qwen 2.5 to KerasHub. This allows developers to leverage Qwen 2.5’s NLP capabilities across TensorFlow, PyTorch, and JAX backends.

Melatih Vision Transformer dengan Keras 3 by AI GDE Muhammad Ghifary (Indonesia), written in Bahasa Indonesia, explains how to implement and train a ViT model using Keras 3.

Uncontrolled Illegal Mining and Garimpo in the Brazilian Amazon co-authored by AI GDE Maria Luize PInheiro (Brazil) is a paper published on Nature Communications, using deep learning to detect iIllegal mining in the Brazilian Amazon with Google Earth Engine, Keras, and Tensorflow.

Vertex AI & Cloud

Prompt Optimizer Made Easy with Vertex AI in GCP

Prompt Optimizer Made Easy with Vertex AI in GCP by AI GDE Esther Irawati Setiawan (Indonesia) guides how to optimize and improve the performance of prompts using Vertex AI Prompt Optimizer.

Automating Hoax Classification with Vertex AI and Cloud Endpoints by AI GDE Esther Irawati Setiawan (Indonesia) shows how to implement a classification system that can classify claims as hoaxes, verified information, disinformation, or hate speech. It covers from training of the model using Vertex AI to its deployment through Vertex AI Endpoints as well as evaluation of the model’s performance and accuracy.

Creating an AI Travel Blog Assistant to Document Patagonian Adventures (Colab notebook) by AI GDE Lesly Zerna (Bolivia) shows how she built a tool that analyzes travel photos and generates detailed descriptions, historical context, and travel recommendations. She used Vision AI, to identify locations and geological formations, and Gemini to create content for blogs and social media.

Vertex AI Express Mode and LangChainJS by AI GDE Allen Firstenberg (US) introduces how to use Express Mode for easy access to Gemini and use it with LangChainJS.

[Python] Replace Gemini with Vertex AI in LangChain by AI GDE Evan Lin (Taiwan) shows how to migrate Gemini to VertexAI using a chatbot as a sample.

Google Cloud Model Armor (repository) by AI GDE Sascha Heyer (Germany) explains why you need Model Armor and how it works. He walks you through how to set up to use it and the limitations and considerations of it.

Google Cloud Model Armor (repository) by AI GDE Sascha Heyer (Germany) explains why you need Model Armor and how it works. He walks you through how to set up to use it and the limitations and considerations of it.

LINE Messaging API x Vertex AI Search: Bringing Search Engine Capabilities to LINE Chatbots by AI GDE Punsiri Boonyakiat (Thailand) explores how to use Vertex AI Search for advanced data retrieval and to create a LINE chatbot with a search engine inside that can interpret and analyze text and images.

How to Integrate AI Models with the Vertex AI API: The Ultimate Developer Tutorial (Colab notebook) by AI GDE Carlos Alarcon (Colombia) is a step-by-step guide to get started with Vertex AI and connect Gen AI models like Gemini to your projects.

Building a Real-time AI Music Sentiment Analysis System by AI GDE Khongorzul Munkhbat (Mongolia) shows how she built a music mood classification system using Vertex — from data preparation to deployment as a web application that anyone can use.

Imagen 3: Beyond Image Generation by AI GDE Margaret Maynard-Reid (US) was a talk exploring the latest features of Imagen 3 — from generating and editing images to fine customization of images. You can see the short video here.

ODML

flutter_gemma 0.8.4 by AI GDE Sasha Denisov (Germany) is a plugin that enables running Gemma directly within Flutter apps making on-device LLM integration simple and efficient. As this was designed to bring modern AI capabilities to Flutter apps with minimal setup, it also supports other SLM as well.

Fine-Tuning Gemma with LoRA for On-Device Inference (Android, iOS, Web) with Separate LoRA Weights (Colab notebook) by AI GDE Sasha Denisov (Germany) is a step by step guide to finetune Gemma-2B using your own dataset and LoRA. The goal is to enable AI applications that run directly on mobile devices or in browsers, ensuring privacy and reducing reliance on cloud infrastructure. He also gave a talk, Building Mobile and Web Apps with GenAI capabilities without Connection, Money and Privacy Concerns to help mobile/web developers to enhance user interaction and functionality with Gemma.

Community Spotlight

BuildWithAI: Gurugram Edition by TFUG Ghaziabad covered in-depth and practical topics. AI GDE Bhavesh Bhatt (India) delivered a session on Graph RAGs (slides). AI GDE Nitin Tiwari (India) shared on VLM and Gemma 3 (slides). Cloud GDE Ashutosh Bhakare (India) led a hands-on codelab on Using Vision APIs with Python (slides | Colab notebook).

Build with AI: Gemini Hands-on Workshop by Deep Tech Stars was a hands-on workshop covering how to make a simple chatbot using Gemini (with this codelab) and how to generate images programmatically using Imagen 3 (slides | Colab notebook).

Build with AI — Workshop by AI Practitioners Chennai was on two practical hands on — 1) building an AI powered fashion app with AlloyDB and Gemini 2.0 and 2) building contextual yoga poses recommender with Firestore, Vector Search, and Gemini 2.0.

AI Day Day: Build With AI by ML Bhopal was one of the biggest events with 400+ participants. AI GDE Anubhav Singh covered the first session, Gemini as a Judge for RAG Evaluation and Cloud GDE Jitendra Gupta led a hands-on workshop on getting started with Vertex AI Studio. From tech talks to networking and a showcase of AI-driven products or solutions, this event covered various topics of the latest AI developments including AI agents.

Build with AI Indore — Workshop 01 by ML Indore shared how to design, build, and deploy a web application that recommends yoga poses using Vector Search and AI-powered tools using Gemini and LangChain.

[Mar 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.