[Nov] ML Community — Highlights and Achievements

[Nov] ML Community — Highlights and Achievements

Let’s explore highlights and accomplishments of the vast Google Machine Learning communities over the month. We appreciate all the activities and commitment by the community members. Without further ado, here are the key highlights!

Featured Stories

NeurIPS 2024 Blog by AI/ML GDE Chansung Park (Korea) is an auto-generated blog that provides summaries of papers in the AI field. This blog is powered by his another project, Paper Reviewer, which turns arXiv papers into blog posts along with audio podcasts. He used Gemini and Vertex AI to generate summaries, in-depth insights, better captions for figures and tables, and podcast scripts.

The Power of Advancement Showcases Achievements | [GDE Sharing] Reflections on Fifteen Years of Growth Alongside Google (Chinese version) by AI/ML GDE Yucheng Wang (China) is a retrospective post celebrating his 15-year journey with the Google Developer community. He focuses on his growth and development within Google’s community programs emphasizing the value of continuous learning, collaboration, cross-cultural communication and interaction with Googlers.

AI/ML Community Summit 2024

AI/ML Community Summit, the annual gathering of the most active members of AI/ML GDEs and community organizers was held in November 2024. Participants shared their activity highlights, feedback to Google’s programs and products, and also got product updates from Googlers in the Gemma, JAX and Keras teams.

Product Highlights

Gemini

How to use generative AI to create podcast-style content from any input (repository) by AI/ML GDE Dimitre Oliveira (Brazil) is about how to use Gemini and Cloud Speech API to create a DIY version of the NotebookLM product.

CosmoGemini (repository) by AI/ML GDE Bhavesh Bhatt (India) won 1st Prize in the Professionals Track of Cosmocloud Low-Code Hackathon. His solution used Gemini Pro 1.5 to create an enterprise RAG solution using Cosmocloud backend no-code solution.

Build a prototype book inventory app in a single day by AI/ML GDE Daniel Gwerzman (UK) is a video about how he used Gemini’s multimodal capabilities to build an app that recognizes book titles from the photos of book shelves.

https://medium.com/media/1f19faf493ec6609f4f1cba427ec2c17/href

FlashBack — Recall in real time! (repository) by AI/ML GDE Xiaoquan Kong (China) and his team won the Duke University’s AI Hackathon Education Track with a Gemini-powered app. It helps you refresh your memory by recording and analyzing footage from your point of view.

Plato Assistant- Your Baby Cooking Companion by AI/ML GDE Xavier Portilla Edo (Spain) is an Gemini-powered app designed to help you cook nutritious meals for babies of all ages. It simplifies the process of planning, preparing, and cooking meals tailored to your baby’s specific dietary needs.

Gemini Docker Service Log Analysis by AI/ML GDE Linda Lawton (Denmark) explains how to embark on a comprehensive analysis of web server log files generated within Docker containers by leveraging Gemini’s exceptional long context window capabilities. This approach enables you to uncover hidden patterns, identify root causes of issues, and proactively address potential security threats.

Implementation of REAcT Agent using LlamaIndex and Gemini by AI/ML GDE Tarun R Jain (India) dives deep into REAcT Agent, a powerful approach in agentic workflows from the paper released by the Google Research and Brain team along with Princeton University. He explores REAcT prompting, why it’s useful, and how to implement it using LlamaIndex and Gemini LLM.

[ GenAI Lab ] 用 OpenAI Python SDK 來使用 Google Gemini Model API

Use the OpenAI Python SDK to access the Google Gemini Model API (Colab Notebook 1, 2) by AI/ML GDE Liu Yu-Wei (Taiwan) explains how to use the Gemini API through the OpenAI Python SDK. It provides a step-by-step guide to setting up and using Gemini’s capabilities, including obtaining an API key, connecting through the OpenAI library, and running sample code to make API calls. He also shared another article about an AI agent using Gemini and Langchain to analyze the results of 2024 WBSC Premier 12 (Colab Notebook).

Build GenAI apps With Chrome built-in APIs (repository | demos) by Angular GDE Connie Leung (Hong Kong) introduced the experimental Chrome built-in APIs and demonstrated 3 demos using Prompt, Translation, Writer, Summarization, and Language Detection APIs.

Google’s Gemini API for Music Composition (slides) by Web GDE Sebastian Gomez (Colombia) was a talk about the tuned models in Gemini for music composition. It shows how to use Gemini API and AI Studio to compose and tune music based on theory.

Legal AI Assistant for India’s New Laws | Gemini Long Context (Kaggle Notebook) by AI/ML GDE Rishiraj Acharya (India) is a video introducing his recent project leveraging Gemini 1.5’s long context window. The notebook explores how the 2-million-token capacity of Gemini 1.5 allows it to process the thousands of pages long legal documents while analyzing case-specific facts.

Google Gemini is Now Smarter Than Ever! Check Out Its New Search Feature by AI/ML GDE Carlos Alarcon (Colombia) introduces the new feature in Gemini, “Google Search”. He explains how it helps Gemini by connecting the model directly to Google’s search engine without any intermediaries.

Empower your AI application with Real-Time Insights by AI/ML GDE Linda Lawton (Denmark) shares how Google Search Grounding helps AI responses be accurate and up-to-date.

Hands on Gemini API and VertexAI (Colab Notebook) by AI/ML GDE Lesly Zerna (Bolivia) was a workshop at DevFest Santiago that led the participants to get their Gemini API key and build their first line of code in Colab to use Gemini Flash to start building applications.

Gemma

LLMinABox: On Device Personalized Diary And Concierge using your voice and Gemma (slides) by AI/ML GDE Rabimba Karanjai (US) is an AI-powered solution that turns your device into a personalized assistant. Leveraging Gemma and on-device processing, it provides voice interaction and intelligent assistance by transcribing and organizing your recordings, summarizing meetings, highlighting key points, and automatically extracting actionable items.

We all have the right to be spam free — Smart Spam Filter developers (slides) by AI/ML GDE Junbum Lee (Korea) shared his experience developing the Smart Spam Filter application for iOS using on-device LLMs models including Gemini Pro. His app received 4.6 stars ranking #2 in the iOS utility section. He also shared Gemma-Ko: Bringing Korean Language Capabilities to Open Language Models focusing on enhancing the performance of Gemma to support Korean language understanding and response generation.

Speculative Decoding: A Guide With Implementation Examples by AI/ML GDE Aashi Dutt (India) is a tutorial guiding how speculative decoding works, when to use it, and how to implement it using Gemma 2.

Intro of ShieldGemma by AI/ML GDE Jerry Wu (Taiwan) talked about the challenges facing LLMs today. He explained how ShieldGemma can be used to protect language models from generating harmful content or hate speech. He covered technical content as well as case studies of the model.

2024 AI/ML workshop materials — Gemma by AI/ML GDE Yucheng Wang (China) is a list of Colab notebooks he used for the entire 2024 China DevFest season. Hundreds of participants have been trained with these notebooks.

gemma-cookbook by AI/ML GDE Xiaoxing Wang (China) is the Chinese translated version of the collection of guides and examples for Gemma.

Vision Language Models & PaliGemma at DevFest Lucknow 2024 by AI/ML GDE Nitin Tiwari (India) was a talk about VLMs, their architecture, and PaliGemma’s ability to handle different tasks. He also showed examples of how PaliGemma can be used for object detection and segmentation.

Full-stack development and local deployment of LLM App based on Gemma by AI/ML GDE Wei Zheng (China) introduced how to use Gemma and PaliGemma to achieve multimodal interaction, and combine RAG to achieve knowledge injection and tool invocation. He guided how to use Flutter to build cross-platform UI to interact with LLM App.

Responsible AI

https://medium.com/media/a9c881f687a668a6f1e2b6cb6592ebef/href

AI/ML GDEs Daniel Goncharov (US) and Jigyasa Grover (US) participated in the Responsible AI panel at DevFest Silicon Valley 2024 and discussed their personal journeys in technology and explored the future of AI.

A quick tour into Google’s Secure AI Framework (SAIF) by AI/ML GDE Patricia O’Callaghan (UK) explored SAIF, an essential guide to enhancing AI security. This session broke down SAIF, examining its core elements that address AI-specific threats, promote security by design, and encourage industry-wide collaboration.

Federated Learning at Scale: How Gboard Privately Serves Billions with AI AI/ML GDE by Konstantinos Kechagias (Greece) was a talk that introduced the fundamentals of Federated Learning and highlighted its real-world application in Gboard. It touches on how secure multiparty computation and differential privacy allow Gboard to enhance its AI models while preserving user privacy, offering a glimpse into the challenges and breakthroughs of deploying this technology at scale.

Responsible AI and Gemma Scope by AI/ML GDE Xiaohu Zhu (China) explored Google’s Responsible AI approaches and related RAI toolkit. He also introduced Gemma Scope which is a mechanistic interpretability tool for LLMs and showcased Gemma Scope examples.

Keras

Using Keras for Gemma model inference and fine-tuning by AI/ML GDE Wei Zheng (China) was a workshop to guide the steps for fine-tuning and inference Gemma 2 using Keras.

No PhD Required: IA Generative AI with Keras (Colab Notebook) by AI/ML GDE Ahirton Lopes (Brazil) aimed to show how Keras, with its high-level API, break down the barriers to developing complex models, making generative AI accessible without deep academic knowledge.

JAX

JAX things to watch for in 2025 by AI/ML GDE Grigory Sapunov (UK) introduces 7 experimental things that tend to become the mainstream, including NNX and shard_map.

https://medium.com/media/b565083d432bbc9943d7874fb93148a2/href

Metric Self-supervised Learning with Jax/Flax by AI/ML GDE Taha Bouhsine (US) explores various self-supervised learning methods, such as SimCLR and BYOL, and help you choose in implement the method that fits your requirements and data using JAX/Flax or TensorFlow.

Firebase Genkit

https://medium.com/media/f4a81df93ca295d9f2b849ae9f3f19d0/href

Introduccion a Firebase Genkit con Github Models (video) by AI/ML GDE Xavier Portilla Edo (Spain) guides how to deploy Genkit’s open-source applications for Node.js and Go.

Super power your app with Genkit by AI/ML GDE Pankaj Rai (India) was a talk at AI Day hosted by TFUG Bhubaneshwar. He presented various platforms offered by Google to get started with Gemini and Genkit features with advantages.

VertexAI

Ampy: An advanced green energy assistant, Built with the Gemini API by AI/ML GDE Konstantinos Kechagias (Greece) helps you to choose the right energy provider and provides personalized data-driven energy saving advice. The video explains how he built it and leveraged VertexAI.

Grounding with Gemini using Vertex AI Search by AI/ML GDE Lai Fong Leong (Malaysia) introduced Grounding in Vertex AI, the capability that lets the model access information at runtime that goes beyond its training data. She guided how to get more accurate, up-to-date, and relevant responses from LLMs.

Others

Create your own INFINITY STONE GAUNTLET DEMO! using TensorFlow (repository) by Machine Learning, AI, Deep Learning & NLP Community — Bangladesh. In the session, speaker Zaynul Abedin Miah, demonstrated how to use TensorFlow.js for a hand pose recognition project. The project provides real-time recognition using your webcam and you can check it out on demo.

https://medium.com/media/affd068dbb5ce971576e218122149b62/href

ML Math Clubs Sucre — video 1, 2, 3 by GDG Sucre was a month-long campaign covering basic concepts of ML and math concepts such as descriptive statistics, linear regression, evaluation metrics, and cross-validation, combining theory with practical implementations in Python and scikit-learn.

[Nov] ML Community — 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.