[Jan 2025] AI Community — Activity Highlights and Achievements

[Jan 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/4894d4aa8e1e30b4123232b3e08c3103/href

Geminar: Interactive Seminars with Real-time Feedback and Gemini based Q&A (repository) by AI GDE Rishiraj Acharya (India) is a tool designed for interactive seminars with real-time feedback and Q&A management system using Gemini. Inspired by Dylan Wiliam’s colored cups concept, he implemented features that allow attendees to show their understanding in 4 degrees so that the presenter can decide whether to speed up or slow down the learning process and when to handle questions.

Why to choose Gemma as an open AI model

Why to choose Gemma as an open AI model by AI GDE George Soloupis (Greece) is a blog post that introduces Gemma, focusing on the training techniques & data preprocessing methods used for this model, its performance, safety & ethical considerations, and real-world use cases. This article provides readers with a better understanding of Gemma and reasons why Gemma is a safer choice for developers concerned about responsible deployment and potential risks of less transparent LLMs.

Product Highlights

Gemini

https://medium.com/media/caa8df0c07919c0ddef834b85abd13a6/href

Gemini 2.0 Flash by AI GDE Sam Witteveen (Singapore) introduces the latest updates of the model and how it can do various multimodal tasks such as generating output in audio/images and interacting in different languages naturally in real time. He also shows demo of the new experimental model that’s trained to generate its “thinking process” in Gemini 2.0 Flash Thinking (Colab Notebook).

Gemini 2.0 — How to use the Live Bidirectional API by AI GDE Sam Witteveen (Singapore) takes a look at the bidirectional API for Gemini 2 that allows you can have live conversations and video interactions with the model.

https://medium.com/media/4ae625525caab62ceabd084a9b9e9e38/href

Gemini 2.0 Flash: Real-Time Multimodal AI for FREE! (Spanish) by AI GDE Carlos Alarcon (Colombia) is a video introducing the “Stream Realtime” feature in Gemini 2.0 in AI Studio. He shows how it works in action with instant responses through audio, video, and text.

Grade images with Gemini 2.0 Flash (repository) by AI GDE Sayak Paul (India) is an online demo of his script that shows how to use the model to grade images based on various aspects of a prompt.

Gemini 2.0 Flash in Action: AI Learning Buddy Demo (repository | demo) by Zaynul Abedin Miah (Bangladesh) is an AI-powered learning tool creating stories and illustrations on the topic requested. It has adaptive difficulty levels, quiz generation, and learning progress tracking features.

Grounding Results with Google Search, Gemini, and LangChainJS

Grounding Results with Google Search, Gemini, and LangChainJS by AI GDE Allen Firstenberg (US) shows how Grounding with Google Search works, and how you can enable it in your calls to Gemini.

Grounding LLMs by AI GDE Juan Guillermo Gomez Torres (Mexico) shares what Grounding is, how it works, how to use it with Vertex AI, and use cases.

Giving Gemini Google Search Access (Colab Notebook) by AI GDE Sam Witteveen (Singapore) goes through Grounding with Google Search and shows how you can improve the responses and results that you get back from Gemini.

Build a Multimodal Agent for Product Ingredient Analysis by AI GDE Tarun R Jain (India) walks you through building a Product Ingredients Analyzer using Gemini 2.0, Phidata, and Tavily Web Search so that you can learn quickly about the chemical ingredients in the products on your hands.

Saving Time and Money using Gemini with Batch Prediction by AI GDE Juan Guillermo Gomez Torres (Mexico) introduces benefits of batch prediction, its use cases, and how to use it in Vertex AI. You will learn how to save time and money by leveraging batch prediction, instead of processing individual requests to Gemini.

Open-source implementation of OpenAI’s Tasks feature (repository) by AI GDE M. Yusuf Sarıgöz (Türkiye) is a Gemini version of Tasks feature which schedules tasks and makes reminders.

RAG & Roll: Rocking Your Enterprise Data with AI on Vertex AI (slides) by AI GDE Svetlana Meissner (Germany) explores how to transform enterprise data into a powerhouse of insights. She covered various techniques RAG, how to use it to enhance accuracy and relevance of your app, and how to use Grounding with Google Search, Vertex AI Vector Search, and Vector Embeddings, etc.

LLM Apps: Evaluation

LLM Apps: Evaluation course (repository) by AI GDE Ayush Thakur (India) is a free online course on LLM Applications evaluation. This course features Gemini as LLM Validators and leverages the generative-ai-python library for making LLM calls. It was created in collaboration with Googler Paige Bailey and Graham Neubig.

Gemini with Function calling in Node by AI GDE Vrijraj singh (India) is a codelab walking you through how to use Gemini’s function calling capabilities. You’ll create an API that seamlessly converts Indian Rupees to US Dollars using real-time exchange rates. This hands-on project demonstrates how to integrate NLP processing with programmatic functions through Gemini’s API. He also shared another codelab, The Meme Stack: funny-tech-memer × Gemini guiding you how to build a custom meme generator using Gemini API & JavaScript.

https://medium.com/media/4408624cb134daa30e5b6758c2f598bb/href

WP On-Device AI Assistant (video) by AI GDE Nico Martin (Switzerland) is a WordPress plugin that provides an on-device AI assistant with features that help your writing by rewriting paragraphs and generating titles, etc. This project was submitted to the Google Chrome Built-in AI Challenge and he won the “Best ‘Real-World’ App (Web Application)” award. BenzGPT and Ask my Website are also his projects that use Prompt API and work on your browser.

Build a sentiment classifier with Chrome’s Prompt API in Angular (repository | demo) by Angular GDE Connie Leung (Hong Kong) walks you through how she built a sentiment classifier using Chrome’s Prompt API, N shots prompting, and Angular. She passed some examples on to the prompt session so the Gemini Nano knows about positive and negative emotions. She also shared Configure temperature and topK in Chrome’s Prompt API (repository | demo) which configures the temperature and topK of a prompt session so that Chrome’s Gemini Nano can generate creative answers at a higher temperature.

Gemma

Vision Language Models & PaliGemma 2 (slides) by AI GDE Nitin Tiwari (India) was a beginner-to-intermediate friendly talk covering architecture and the working of Vision Language Models. With a deep dive into PaliGemma and its ability to handle various tasks, he showed hands-on examples and also talked about its enhanced capabilities for transfer tasks such as table structure recognition, chemical molecular structure recognition, etc.

Gemma2 Fine-Tuning: From SFT and QLoRA to GGUF Deployment with Ollama by Byunggil Yoon (Korea) shows how to fine-tune Gemma2–2b-it, using tools like TRL, Transformers, Datasets, and PEFT. He also used supervised fine-tuning and direct preference optimization techniques to create a Korean QA bot.

ML Research

[Future Trend] AI Agent: Experience — Learn more about AI Agent from the eyes of Google experts by AI GDE Liu Yu-Wei (Taiwan) is a review and summary of the white paper, Agents by Google.

AI Agents explained: When models learn to plan and act by AI GDE Patricia O’Callaghan (UK) was a session exploring how AI agents combine reasoning, tools, and decision-making to execute real-world actions. She also discussed the recent white paper published by Google.

Community Highlights

Turing’s Class: The Math That Powers Machine Learning (Youtube playlist) by Team TechnoJam was a part of the ML Math Clubs activities sharing the fundamental mathematical concepts needed to understand and build ML models. Focusing on topics such as linear algebra, calculus, probability, and advanced math techniques, participants engaged with the core topics in an interactive, hands-on format. ML Math Clubs: Strengthening Machine Learning Foundations, One Session at a Time is a blog post that summarizes their campaigns, achievements and future plans.

ML Mumbai December Event by TFUG Mumbai was a blend of learning and hands-on experience covering Gemini 1.5 Flash and Pro, as well as the latest Gemini 2.0 and Gemma. Participants learned their capabilities to chat, create, analyze, build, and integrate into real-time projects via the Gemini API and AI Studio.

ML Study Jams (a series of 8 sessions) by ML Community Talk Agadir was a beginner-friendly campaign on topics ranging from Python programming to CV and responsible AI. This energetic series fostered community engagement campaigns through collaboration with a local GDG on campus.

[Jan 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.