![[feb-2025]-ai-community — activity-highlights-and-achievements](https://prodsens.live/wp-content/uploads/2025/03/32806-feb-2025-ai-community-activity-highlights-and-achievements-550x371.png)
[Feb 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/95bae902f625262e66471ab60308f9cf/href
Gemma 3 — The NEW Gemma Family Members Have Arrived!!! by AI GDE Sam Witteveen (Singapore) is a video looking at the release of the new Gemma 3 model in four sizes — 1B, 4B, 12B, and 27B.
https://medium.com/media/a00829b4cb426f9de65f9ee1a3981de8/href
Gemma 3 Function Calling with Open Source LLMs and Building a Local Agentic RAG System using Gemma-3, FastEmbed and Qdrant Vector database by AI GDE Tarun R Jain (India) explain and explore the new features of Gemma 3.
https://medium.com/media/2f1155fdbbf7ef65f0cde5c6223cb6ba/href
Gemma 3 For Video Understanding (Colab Notebook) by AI GDE Merve Noyan (France) is a notebook on interleaving frames and doing video inference, showing Gemma’s video understanding capabilities.
Inference Gemma 3 (Colab Notebook) by AI GDE Nitin Tiwari (India) is a notebook that demonstrates Gemma 3’s vision and multilingual capabilities on images and videos.
https://medium.com/media/f24c9f31d82db0c918485806561c31e8/href
Unleash the Power of AI: Host Your Own Ollama Models for Free with Google Colab! (video | Colab Notebook) by AI GDE Dimitre Oliveira (Brazil) is a guide on how you can use a Colab instance as a remote server to host your models and query them from various sources. He also used the new Gemma 3 models with Ollama.
Community Spotlight with JAX/Flax
Building Scalable Models In JAX & FLAX (Colab Notebook) by AI GDE Henry Ndubuaku (UK) at Machine Learning Lagos event was a walk-through on building scalable ML models using JAX/Flax. The session provided a comprehensive introduction to JAX, highlighting its powerful JIT compilation capabilities and seamless integration with the XLA Compiler to boost performance.
Vertex AI & Gemini
Content Creator AI Tools (video) by AI GDE Dimitre Oliveira (Dimitre) is an app powered by Gemini (and Vertex AI) to help you create technical content. It can take your code repository, text input, files, etc. to transform them into content like blog posts, Github README files, improved codes, videos and more. Additional blog post, Supercharge Your Technical Content Creation with Gemini, will also walk you through the process.
https://medium.com/media/dd1c1c5ba49a7a90668331883d4ea1a6/href
Use #vertexai #gemini 2.0 Flash + #langgraph to generate character backgrounds and game stories through #game Sprite (repository) by AI GDE Cyrus Wong (Hong Kong) is about how he generated RPG characters and background stories with Vertex AI and Gemini 2.0.
Better PDF OCR in RAG: Gemini 2.0 as a LangChain Loader (Colab Notebook) by AI GDE Daniel Gwerzman (UK) explores how Gemini can extract text more effectively, especially when dealing with complex layouts, including images. He also compared Gemini’s output to existing PDF parsers, highlighting the strengths and weaknesses of using an LLM-powered OCR for structured text extraction.
A Personal Assistant for knowledge management based on Gemini 2.0 and Vertex AI (Colab Notebook) by AI GDE Luca Massaron (Italy) is a tutorial on using Vertex AI and Gemini to create a tool to extract key information from web pages, generate embeddings for tags, order bookmarks, and generate a markdown document with extracted data.
Evaluating an RAG in LLM using Vertex AI
Evaluating an RAG in LLM using Vertex AI (Colab Notebook) by AI GDE Joan Santoso (Indonesia) discusses methods for evaluating RAG within LLMs using Vertex AI Evaluation API. It emphasizes the importance of evaluating both the retrieval and generation components of RAG systems to ensure accurate and contextually relevant outputs. The author highlights tools and services provided by Vertex AI.
Multi-Agent Content Creation with Gemini & Crew AI (Colab Notebook) by AI GDE Nathaly Alarcon Torrico (Bolivia) guides how to create AI agents with Gemini and Crew AI. Based on the scenario of a digital agency for content creation, she created 4 agents in each role, supervisor, researcher, writer and coordinator, and had them research on “AI in Latin America” and create marketing content.
Build an Xtreme Weather App with Google Geocoding and Places API by AI GDE Rubens Zimbres (Brazil) is a weather monitoring and disaster alert system built with LangChain and Gemini 2.0 Flash. The app takes an user’s address as input and provides personalized emergency guidance by combining real-time data from OpenWeatherMap, USGS and Google Maps APIs so that a user can prepare for a disaster like flood, hurricane, tsunami, etc.
Understanding Alzheimer’s: Building Knowledge Graphs from Unstructured Data with Gemini
Understanding Alzheimer’s: Building Knowledge Graphs from Unstructured Data with Gemini by AI GDE Rubens Zimbres (Brazil) shows how he used unstructured PDF data about Alzheimer’s to build a Knowledge Graph with the help of Gemini 1.5 Flash and Neo4j. It’s to better understand the disease, its causes, effects, and possible treatments at the gene level and protein level.
AdaptSum by AI GDEs Sayak Paul (India) and Chansung Park (Korea) is an LLM-powered system for dynamic summarization. Instead of generating entirely new summaries with each update, the system intelligently identifies and modifies only the necessary parts of the existing summary. This post, Distilling from Dialogues: Finding Meaning in LLM Interactions, shows details of the project.
AI Agents by AI GDE Sascha Heyer (Germany) introduces how to make an AI agent using Vertex AI Engine that can integrate Gemini models, Python functions, and tools like LangChain for orchestrated reasoning. He starts from understanding AI agents and covers building and deploying an agent and debugging issues.
Vertex AI Search App to search results from Angular and Vue 3 documentation sites by Angular GDE Connie Leung (Hong Kong) explains how she built an app searching web pages like the official Angular documentation so that you can focus on the project itself on the development process.
News Article Extractor & Analyzer With Firecrawl, Lamatic and Gemini by AI GDE Vrijraj Singh (India) is a codelab that guides you through building an AI-powered system that scrapes news articles from websites, processes them with AI, and analyzes their content.
EduLink AI (repository | tutorial) by AI GDE Geeta kakrani (India) is a learning assistant leveraging Gemini and Vertex AI to generate summaries and quizzes using PDFs that you provide.
Vertex AI & Imagen 3
Build 3D Reconstructs using Imagen 3 and Stable Video 3D (Colab Notebook) by AI GDE Nitin Tiwari (India) is a project transforming your ideas into 3D visuals. It demonstrates the process of creating 3D objects by first generating images with Imagen 3, then converting those images into 3D meshes using the Stable Video 3D (SV3D) model.
Travel in Style, Rain or Shine🧳Smart Suitcase Packing with Gemini 2.0, Imagen 3, and Open Weather! (repository) by AI GDE Jigyasa Grover (US) leverages Gemini 2.0 and Imagen 3 to create personalized packing lists with the information about when, where, and how long you will travel in what weather. It also generates images of suggested outfits combinations based on your style preferences.
Gemma
Welcome Gemma 3: Google’s all new multimodal, multilingual, long context open LLM
Welcome Gemma 3: Google’s all new multimodal, multilingual, long context open LLM co-authored by AI GDEs Aritra Roy Gosthipaty (India), Merve Noyan (France), and two Hugging Face engineers introduces the newly released Gemma 3 and its enhanced capabilities.
PaliGemma 2 Mix — New Instruction Vision Language Models by Google (demo) by AI GDEs Aritra Roy Gosthipaty (India), Merve Noyan (France), and Googler Andreas P. Steiner is about PaliGemma 2 Mix helping people know what these models are good for, how they differ from the normal PG 2 models! Aritra and Merve also contributed to SigLIP 2, that outperforms the older SigLIP ones and co-authored SigLIP 2: A better multilingual vision language encoder with their fellow co-worker Pavel Iakubovskii (Hugging Face).
PaliGemma 2 Mix: A Guide With Demo OCR Project by AI GDE Aashi Dutt (India) is a tutorial to build an AI-powered bill scanner and spending analyzer extracting and categorizing expenses from bills. It performs OCR to retrieve key information and summarizes spending based on the input images.
Creating your own private Gemma API with Google Vertex by AI GDE Hannes Hapke (US) explored the practical implementation of Gemma on Vertex AI, providing a comprehensive walkthrough of the deployment process. He covered techniques for creating private LLM instances, ensuring your sensitive information remains protected throughout the inference pipeline.
Gemma Fine-tuning for En-Ru medical translations by AI GDE Grigory Sapunov (UK) is a guide to create a Machine Translation (MT) system for English-Russian medical translations in the ophthalmology domain. This system is based on the Gemma 2 Instruct 2B model. He also shared Ophthalmology Russian/English Translations dataset.
Training for Reasoning with GRPO
Training for Reasoning with GRPO — part I & part II by AI GDE Luca Massaron (Italy) demonstrate how to train Gemma 2 2B-IT model to successfully solve high school-level math problems using GRPO , a method recently introduced by the DeepSeek team and implemented in the Hugging Face TRL package.
ODML
Inference PaliGemma 2 with Transformers.js (repository) by AI GDE Nitin Tiwari (India) is a post showing different approaches where you can use Gemma in web apps. Part 1 of the post covers the detailed process of converting and quantizing the PaliGemma 2, which includes the SigLIP vision encoder, Gemma 2 language decoder, and embedding tokens to ONNX format. In part 2, he demonstrates how to use Hugging Face Transformers.js to run the model in the browser for tasks like zero-shot object detection, image captioning, OCR, and visual Q&A.
Fine-tune Gemma for any language on Vertex AI and deploy it to Android by AI GDE George Soloupis (Greece) is a tutorial on how to finetune Gemma for any language and converting the finetuned model directly into a .gguffile within the Vertex AI environment for easy deployment on Android devices.
The power of on-device AI: full privacy with Gemini Nano by AI GDE Gerard Sans (UK) was a talk about how to utilize Gemini Nano on your device or browser, while staying off the network, enabling secure and private AI apps.
Firebase Genkit
Food Fit (repository | tutorial) by Luis Eduardo is a web application where you can search for healthy recipes. The app provides detailed information about food such as ingredients, expected nutritional content, and reference image of the dish, which was created by Imagen 3. At the backend, everything is orchestrated with Genkit, using the schemas to return a structured JSON and Cloud Functions to deploy the APIs that are consulted from the frontend with Angular.
ML Research
tt-scale-flux by AI GDE Sayak Paul (India) is a re-implementation of Inference-Time Scaling for Diffusion Models beyond Scaling Denoising Steps.
Machine Learning for Tabular Data
Machine Learning for Tabular Data — XGBoost, Deep Learning, and AI by AI GDE Luca Massaron (Italy) and Googler Antonio Gulli is a newly published book. It covers how to use XGBoost and LightGBM on tabular data, optimize deep learning libraries, and use cloud tools like Vertex AI to create an automated MLOps pipeline.
[Feb 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.
