Module 2: Introduction to Generative AI and LLMs

Module 2: Introduction to Generative AI and LLMs#

Module Objectives#

This module provides foundational knowledge on Large Language Models (LLMs), covering key concepts such as pretraining, foundational models, and adapting LLMs through fine-tuning. Additionally, the module introduces various open-source and proprietary LLMs currently available on the market.

Module Topics#

  • Introducing Generative AI

    • What is Generative AI?

    • How Gen AI differs from Predictive AI.

    • Brief history of Gen AI.

    • Capabilities of Gen AI and major use cases.

    • Different categories of Gen AI (LLMs, image generators, video generators).

  • Understanding Large Language Models (LLMs)

    • Overview and history of language models—LLMs vs. SLMs.

    • Categories of LLMs: Foundation models and other concepts.

    • Building LLMs: Transformer architecture and sequence-to-sequence architectures.

    • Overview of common LLMs: OpenAI models, Mistral AI, Llama, Gemini, and others.

    • Adapting and customizing LLMs: fine-tuning, pre-training, RLHF.

  • Building and Evaluating LLM Apps

    • Key concepts for LLM apps: prompt engineering, prompt-tuning, vector embeddings, RAG.

    • Ecosystem of commercial and open-source tools for building LLM apps (e.g., LangChain).

    • Customizing LLMs for specific use cases: prompt engineering, RAG, fine-tuning, RLHF.

    • Selecting and evaluating LLMs and LLM apps.

  • Deploying LLM Apps with LangChain

    • Overview of LangChain features and capabilities.

    • Preprocessing and loading data in LangChain.

    • Working with different LangChain agents (e.g., SQL).

    • Deploying LangChain applications (e.g., with Streamlit, WhatsApp, and web apps).

    • Evaluating LangChain apps.

Practical Labs#

  • Lab 1: Demonstration of Building an LLM App with Commercial Tools (OpenAI)
    Since participants won’t have access to a paid OpenAI subscription, the instructor will demonstrate available capabilities for building LLM apps. The lab will include:

    • Exploring OpenAI features using the ChatGPT GUI (paid version), showing functionalities and how to create assistants.

    • Demonstrating a simple RAG-based chatbot using the OpenAI playground.

    • Demonstrating a simple RAG-based chatbot using the OpenAI API.

  • Lab 2: Building LLM Apps Using LangChain (RAG-Based Chatbot)
    Participants will use provided documents to build a RAG-based app with an open-source LLM to query the documents. The output will be a Streamlit app for sharing. Tasks include:

    • Setting up the development environment and installing required packages.

    • Preparing source data (e.g., health documents).

    • Setting up a vector database.

    • Preprocessing documents and loading them into the vector database.

    • Integrating with an LLM (including selecting the LLM).

    • Developing the user interface in Streamlit.

    • Deploying and testing the app.

Case Study#

  • Agricultural Information Q&A System in Malawi
    A RAG-based chatbot deployed in Malawi answers questions from Agricultural Extension workers. This example app uses ChatGPT (OpenAI) integrated with agricultural documents from Malawi and is deployed on WhatsApp.

Assessment#

  • Build an LLM App with LangChain
    Participants will receive a notebook to create an app that answers questions based on their selected website. Additionally, a quiz with five multiple-choice questions will be administered.

Contents