Description:

Gen AI and RAG Primer with Gemini

Understanding Large Language Models (LLMs): LLMs Demystified: We'll begin by defining LLMs and exploring their underlying architecture. Powerful libraries and frameworks form the bedrock for building and training these complex models. We'll delve into how these tools enable developers to manipulate massive datasets and train intricate neural networks that power LLMs. Fine-tuning LLMs for Enhanced Performance: We'll explore techniques to improve LLM performance for specific tasks. This involves adapting pre-trained LLMs using well-established tools and methodologies. We'll discuss common approaches like task-specific training and prompt-based learning, all implemented with industry-standard practices. Embeddings and Vector Stores: Efficient Retrieval: We'll introduce embeddings, a technique for representing text data in a way computers can understand. Libraries and toolkits can be leveraged to create these embeddings. We'll then explore Vector Stores, which enable efficient retrieval of similar text data based on these embeddings.

Retrieval-Augmented Generation (RAG) Model: A Powerful Alternative The RAG Difference: We'll delve into the RAG Model architecture, highlighting its key components. Interestingly, the RAG approach can be implemented using established techniques, potentially leveraging familiar tools and methodologies. This section will showcase how RAG leverages the strengths of both retrieval and generative techniques.

Live Demo with Gemini API (Google AI Studio): Putting Theory into Practice Building a PDF Interactor: We'll embark on a live coding demonstration using the Gemini API within Google AI Studio. This interactive session will showcase the creation of a script that interacts with PDFs, leveraging the power of the Gemini API. Embeddings in Action: Within the live demo, we'll revisit embeddings and Vector Stores. We'll demonstrate how the Gemini API potentially utilizes these concepts under the hood to process and retrieve information from PDFs.

What Attendees Take Away:

• Gain a solid understanding of Generative AI (Gen AI) and its core principles.
• Explore the capabilities of Large Language Models (LLMs) and the powerful tools that enable their development.
• Grasp the concept of fine-tuning LLMs and its implementation using industry-standard practices.
• Learn about embeddings, Vector Stores, and their role in text retrieval with established toolkits.
• Understand the RAG Model and its potential advantages over traditional approaches.
• Witness the practical application of Gen AI concepts through a live coding demonstration using a leading API.
• By incorporating references to established tools, methodologies, and libraries throughout the talk, you can subtly highlight the importance of popular Python frameworks in the field of Gen AI. This approach allows the audience familiar with Python to recognize its prevalence without explicitly mentioning the language itself.

Prerequisites:

A basic understanding of LLMs and RAG, if possible.

Video URL:

https://drive.google.com/file/d/1f3T7eX50dzOo9DQ9B73WZ4gsWwU2GRz3/view?usp=sharing

Content URLs:

https://docs.google.com/presentation/d/19oVKSPMLNua-jPweQy6AWXQvWijj-dTOuEzrdLIByW0/edit?usp=sharing

Speaker Info:

I am a second year student pursuing B.Tech CSE from GGSIPU University. I am an advance Python Developer and AIML Enthusiast. I mostly work on computer vision projects but have a good understanding of LLMs as well. I am an intern for contributing in Open Source GIS Tools under eGovernment Foundations in Dedicated Mentoring Program - C4GT'24. Ex - Intel's AI4Youth Intern.

Speaker Links:

https://linktr.ee/SGCODEX