Gemma in the Lab

Join Juyeong Ji from Google DeepMind for a hands-on session introducing the Gemma model family through the lens of scientific innovation. Juyeong Ji - formerly a PlayStation System Software Engineer at Sony and Technical Solutions Consultant at Google (Stadia), now building the future of AI on the Google DeepMind DevX Team. Key Topics Covered Gemma and Gemini Overview: An introduction to Gemini, Google's primary AI service, and Gemma, an open-weights model optimised for local AI agents and private knowledge bases. The Mechanics of Embeddings: A technical explanation of how embeddings use numerical vectors to capture semantic meaning, enabling text similarity calculations and cross-lingual dataset analysis. Retrieval-Augmented Generation (RAG): How RAG addresses the limitation of static LLM training data by grounding language models in external, up-to-date data sources to improve accuracy. Local Database Demonstration: A practical walkthrough using Anything LLM to parse documents into smaller chunks and generate embedding vectors for localized, secure querying. Defining Agentic AI: An look at autonomous systems capable of using external tools without continuous human input, featuring a case study of an autonomous meme generator. Model Selection Criteria: An analysis of Gemma's underlying architecture and why its out-of-the-box efficiency makes it highly suitable for local, agentic workflows. Links from the video: https://ai.google.dev/gemma/docs/embeddinggemma/inference-embeddinggemma-with-sentence-transformers https://huggingface.co/spaces/google/mood-palette https://dev.to/googleai/building-your-first-rag-system-f0n https://ollama.com/ https://anythingllm.com/ https://lmstudio.ai/ https://blog.google/innovation-and-ai/technology/developers-tools/functiongemma/ https://ragnar.tidyverse.org/reference/ragnar_find_links.html https://www.linkedin.com/company/gdg-for-science/ https://gdg.community.dev/events/details/google-gdg-ai-for-science-australia-presents-applying-deep-learning-in-genomics/ https://developers.google.com/health-ai-developer-foundations/overview