![[aug-2025]-ai-community — activity-highlights-and-achievements](https://prodsens.live/wp-content/uploads/2025/09/39330-aug-2025-ai-community-activity-highlights-and-achievements-550x297.png)
[Aug 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
LangExtract — intros, quick demos, and tutorials for RAG systems
https://medium.com/media/59ef953421cff88b7134200367ec9050/href
[👀86k+] LangExtract — Google’s New Library for NLP Tasks (Colab Notebook) by AI GDE Sam Witteveen (Singapore) [👀32k+] LangExtract: Turn Messy Text into Graph-RAG Insights and [👀20k+] LangExtract + RAG: Smarter Retrieval with Metadata Filtering (repository) by AI GDE Muhammad Farooq (US)Gemini 2.5 Flash Image (nano-banana 🍌) tryouts
[Intro][👏74+] Google Goes Bananas!! by AI GDE Aashi Dutt (India) [Demos][👀60k+] Gemini 2.5 Flash Image is Nano Banana!! by AI GDE Sam Witteveen (Singapore) [Tryout][👏102+] My Experience Using the new Gemini 2.5 Flash Image by Cloud GDE David Regalado (Peru) [Code] gemini-nano-banana by AI GDE Gabriel Preda (Romania)Highlights by products
Agent Development Kit (ADK)
MenoGuide+ (repository) by AI GDE Sara EL-ATEIF (Morocco) and her students, is a Flutter-based mobile application designed to empower women navigating menopause. The project was initially created for the ADK Hackathon using the Gemini 2.0 Flash model, and has evolved to leverage a multi-agent system developed with Firebase Studio, where specialized AI agents for symptom analysis, nutrition, fitness, and mental wellness collaborate to provide a holistic and personalized daily plan.
#Gemini Embeddings Embedding and Its Use for LLM Application (slides) by AI GDE Joan Santoso (Indonesia) demonstrated Gemini-Embedding-001 as a powerful text embedding model for cases like semantic similarity and RAG. He used ADK to utilize this embedding model and create an agent that can use the external document based on this embedding.
Hands-On AI: Building Agents with the Google Agent Development Toolkit (ADK) by AI GDE Jigyasa Grover (US) is a LinkedIn Learning course that teaches how to build and test AI agents using the ADK.
https://medium.com/media/58a05ad5112954a8bef26a19e2c0f168/href
Building and Deployig AI Agents with Google ADK and Vertex AI by AI GDE Muhammad Ahsan Ayaz (Sweden) at TFUG Islamabad was a session on building/deploying AI agents using ADK and Vertex AI.
Getting Started with MCP and Google ADK (repository) by AI GDE Tarun R Jain (India) was a workshop on how to build MCP agents using ADK. He also led Git and GitHub Workshop (repository) showcasing how to build an agent using ADK and Gemini, and how to track it using Git commands.
Build a Travel Agent using MCP toolbox and ADK by AI GDE Gaurav Kheterpal (India) was a hands-on session using this codelab to explore concepts like agentic AI, MCP, and A2A.
Gemini
https://medium.com/media/1058bb06c706a95ebc4c0d2e6751c5ba/href
Gemini Deep Think by AI GDE Sam Witteveen (Singapore) and Gemini Deep Think: Built for the Hardest Problems by AI GDE Muhammad Farooq (US) take a look at the new model and see what it can be used for. They dive deep into how it won the gold medal standard in the International Math Olympiad.
Gemini Embeddings: Complete Guide and Practical Tutorial | Revolutionize Your AI Projects (Colab Notebook) by AI GDE Carlos Alarcon (Colombia) introduces what embeddings are; why they are the basis of modern AI; and the advantages of the new Gemini 001 Embeddings. He also covers the Matryoshka architecture and how it allows you to optimize performance and costs.
[👏81+] Enhanced Guide to Using Prompts in Gemini CLI by Cloud GDE Kanshi Tanaike (Japan) provides an overview of how to utilize prompts within the Gemini CLI. Leveraging a Google Apps Script MCP server, it explores practical examples, including roadmap generation, real-time weather inquiries, and Google Drive file searches.
LLM Observability into Gemini on Google Cloud with OpenLLMetry by AI GDE Anjani Kumar Keshari (India) guides you through setting up LLM observability for Gemini models on GCP using OpenLLMetry. It explains how to view the data in Cloud Trace, enabling performance optimization, debugging, and cost analysis.
Gemini CLI
https://medium.com/media/9b4ba4bf6c0f08cbe07aec2e8ee37fac/href
[👏233+] Building a Gemini-powered data catalog extension for Cursor (video | repository) by AI GDE Hugo Zanini (Brazil) shares how he developed a Cursor extension integrating data discovery and lineage into the IDE, leveraging Gemini for natural language search and insights. It allows developers to search data catalogs, view dataset metadata, and explore interactive lineage graphs directly within Cursor, reducing context switching and improving productivity.The Ghost in the Machine by AI GDE Victor Ashioya (Kenya) is an analysis of Gemini 2.5 Pro’s LaTeX inconsistency problem. The article details the widespread issues with Gemini’s handling of LaTeX, including incorrect formatting, whitespace sensitivity, and indentation problems. It proposes three hypotheses for these failings: token laziness, contextual confusion, and the impact of inconsistent training data.
How I Turned My CV into a Portfolio Website and Deployed it in Hours Using Gemini CLI
[👏53+] How I Turned My CV into a Portfolio Website and Deployed it Using Gemini CLI in Hours by AI GDE Ifeanyi Idiaye (Nigeria) shows how he used Gemini CLI to quickly build and deploy a portfolio website. [👏52+] How to solve the bug — Failed to login. Message: Precondition check failed. by AI GDE Yu-Wei Liu (Taiwan) shares his troubleshooting steps for error in Gemini CLI. The article provides four steps to identify and remove the problematic environment variable GOOGLE_CLOUD_PROJECT.all-smi by AI GDE Jeongkyu Shin (Korea) is a command-line utility for monitoring GPU and NPU hardware across multiple systems. It provides a real-time view of accelerator utilization, memory usage, temperature, power consumption, and other metrics. He shared his experience about how he used Gemini CLI as a coding assistant to launch a commercial-grade project.
CameraCoach: AI-Powered Photography Coach by AI GDE Suvaditya Mukherjee (US) is a full-stack progressive web app that provides AI-powered feedback on photographs. You can upload a photo and receive analysis from Gemini 2.5 Pro Vision on composition, lighting, and framing. This app was written completely by Gemini CLI.
https://medium.com/media/4bc298750e78fc489c2142f9d8131fae/href
AI-Powered GitHub Workflow with Gemini CLI by AI GDE DanielGwerzman (UK) dives into this new tool connected with Gemini CLI GitHub Action announced at Cloud Next 25 Tokyo. He covered how to set it up and use it for task automation, and hands-on experiences and insights.
Gemma
Gemma 3 270M: Tiny but Mighty — First Impressions on Kaggle (Kaggle Notebook) by AI GDE Gabriel Preda (Romania) is a hands-on exploration of the new model, showing how even small-scale LLMs can be fine-tuned and deliver strong results on practical NLP tasks. He also shared, Fine-tune Gemma 3 270M using LoRA for Medical Q&A that details the process of fine-tuning.
Gemma 3 270M: A Guide With Demo Project (Android App) (repository) by AI GDE Aashi Dutt (India) is a tutorial to set up, integrate, and run Gemma 3 270M in an Android app, along with best practices for ensuring its reliability and safety for on-device inference.
Tracing the Architectural Evolution of Gemma by AI GDE Krupa Galiya (India) is an overview article of the architectural evolution of the Gemma family — open-weight, generative AI models: from the foundational decoder-only transformer (Gemma 1) to the multimodal and on-device optimized models (Gemma 3 and Gemma 3n). It highlights key architectural upgrades, performance improvements, and trade-offs in each version, including attention mechanisms, context length, and multimodal capabilities.
The Dawn of Offline AI Agents in Your Pocket
[👏142+] The Dawn of Offline AI Agents in Your Pocket by AI GDE Sasha Denisov (Germany) is a comprehensive guide to building offline AI agents in Flutter using the flutter_gemma plugin. Covers model selection, distribution, inference types (single vs. chat), and advanced features like multimodal AI and function calling.Fine-Tuning Gemma 3n for Image-to-LaTeX Task on Vertex AI (Kaggle Notebook | repository) by AI GDE Gabriel Preda (Romania) details the creation of an ML pipeline using Vertex AI Pipelines, Unsloth, and the Gemma 3n 4B vision-language model to finetune the model for converting images of mathematical equations into LaTeX.
AI GDE Yu-Wei Liu (Taiwan) shared two Gemma projects he’s participated in:
- SEA-LION — Fine-Tuning Project on Southeast Asian Languages and Cultures in the AI Wave addresses the under-representation of Southeast Asian languages and cultural nuances in existing mainstream LLMs (The model used Gemma and Llama, and it was integrated into Model Garden).
- taide/Gemma-3-TAIDE-12b-Chat: A New Open Source Model Fine-Tuned by TAIDE on Gemma 3: an open-source model finetuned by the TAIDE (Trustworthy AI Dialogue Engine) project in Taiwan. The article discusses the model’s license, parameters (12B), Hugging Face availability, and performance benchmarks, highlighting its enhanced capabilities in handling Taiwan-specific knowledge.
On-device AI
[👏50+] Run Gemma and VLMs on mobile with llama.cpp (repository) by AI GDE Georgios Soloupis (Greece) explores running Gemma 3 and VLMs offline on Android using llama.cpp and llama.rn, a React Native binding. It details the technical setup for deploying GGUF-formatted models, enabling text-only and multimodal inference. The app supports real-time streaming, vision-text prompts, and models like SmolVLM2 and Gemma 3 4B.,. showcasing the potential for private, on-device AI.Ollama RAG Demonstration with EmbeddingGemma and Gemma3n by AI GDE Jimmy Liao (Taiwan) is a quick post exploring the multilingual capabilities of EmbeddingGemma and Gemma3n.
JAX
Minho also shared his review article, Google I/O Connect China 2025 from the perspective of an AI GDE.
Colab
Data Immersion with Python using Google Colab and Gemini (Live 1, 2, 3) by AI GDE Vinicius F. Caridá (Brazil) was a free online course designed for beginners to learn the fundamentals of Python for data analysis and data science. All sessions were conducted using Colab as the development environment, and Gemini was introduced to accelerate programming learning and knowledge building. Including live streams as well as private recorded classes, the total views of the course reached more than 163k.
AI in Healthcare FDP at Ramaiah Institute of Technology (Colab Notebook) by AI GDE Tarun R Jain (India) was a workshop demonstrating the current challenges in Healthcare. He focused on bias and human-in-the-loop needs in Gen AI apps. He also showcased a multimodal feature using MedGemma.
Gemma LLM on Kaggle: Step-by-Step Guide to Use Google’s Open LLM by AI GDE Geeta kakrani (India) guides step-by-step on how to use Gemma directly from Kaggle and run it in Colab without the internet.
Keras
AI GDE Harshal Janjani (UAE) contributed to KerasHub by implementing T5Gemma.
AI GDE Hongyu Chiu (Chiu) contributed to KerasHub by implementing DINOv2 (repository).
Firebase Studio
AI GDE William McLean (US) shared step-by-step guides on running the codelabs in the Firebase Studio environment: Exploring the Data AgentVerse with Firebase Studio and Deploying a MCP Server on Google Cloud Run with Firebase Studio.
Cloud
Vertex AI Unlocked: Building Your Application with Gemini (slides) by AI GDE Svetlana Meissner (Germany) was a talk covering the core components of Vertex AI for Gen AI and how to integrate Gemini models into applications.
Cultural Virtual Try-On with Imagen 4 (repository) by AI GDE Nitin Tiwari (India) uses a culturally rooted approach to virtual try-ons. Users select a location, and the app uses the Cloud’s Geocoding API and Imagen 4 to generate traditional clothing inspired by the region’s heritage. The outfit is then applied using the Virtual Try-On API.
Cultural Virtual Try-On with Imagen 4 (repository) by AI GDE Nitin Tiwari (India) uses a culturally rooted approach to virtual try-ons. Users select a location, and the app uses the Cloud’s Geocoding API and Imagen 4 to generate traditional clothing inspired by the region’s heritage. The outfit is then applied using the Virtual Try-On API.
Certification Study Group (playlist) by GDG Ahlen was a “Get Certified” campaign designed to help aspiring professional ML engineers. They covered basic ML concepts, data preparation, model optimization, and MLOps tools within Google Cloud.
Community Spotlight
https://medium.com/media/5ad5b4584aadce10a0c738e95897a30f/href
ML Summer School by Machine Learning, AI, Deep Learning & NLP Community — Bangladesh was a collection of AI campaigns including AI Study Jams, AI Math Clubs, AI Paper Reading Clubs, and Kaggle Community Olympiad. Check out the Math Clubs videos for a deep dive into math concepts for AI.
Kaggle Community Olympiad
The Kaggle Community Olympiad campaign is now open as an evergreen campaign. During the pilot program, 8 communities accomplished the campaign’s goal: ML Community Talks Agadir, Aye Hninn Khine, AI Prayagraj, ML Nashik, AI Durg, The Coding Culture, TFUG Bhubaneswar, Machine Learning, AI, Deep Learning & NLP Community — Bangladesh. Please refer to the previous real-world problem-solving competitions here, as well as the hosted training sessions from TFUG Bhubaneswar (video) and ML Nashik (video).
[Aug 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.
![[pt-br]-functional-interfaces,-a-fundacao-para-a-programacao-funcional-com-java](https://prodsens.live/wp-content/uploads/2022/08/2560-pt-br-functional-interfaces-a-fundacao-para-a-programacao-funcional-com-java-380x250.png)
