Mastering Generative AI: Foundations to Advanced Application

The Generative AI program, participants learn to evaluate LLM accuracy using metrics like precision, recall, and F1 scores. A hands-on segment guides them through setting up their environment, using a PDF to create question-answer pairs, and assessing model responses to enhance accuracy.

This session covers tokenization in NLP and generative AI, exploring word, sub-word, and character tokenization to enhance contextual understanding. Participants practice token-level evaluation, using metrics like precision, recall, F1, and ROUGE scores for performance insights in NLP tasks.

The Generative AI program, participants learn AI text summarization techniques—extractive, abstractive, and hybrid—highlighting time-saving and communication benefits. They explore data logging in structured formats like JSON to track AI interactions, enhancing observability and improvement. A hands-on exercise involves generating and analyzing AI summaries for product reviews, providing practical skills in summarization and data analysis.

The Generative AI program, participants dive into "Multimodal AI," integrating text, audio, video, and images to enhance customer interactions. Through demos of speech-to-text and text-to-speech, they explore challenges like accent variability and noise. A hands-on segment with the Groq API enables participants to create a voice-based AI assistant, applying multimodal AI in real-world scenarios like customer care and accessibility.

"Function Calling in Generative AI" delves into enabling AI to execute tasks like booking tickets or managing finances, moving beyond text generation. Participants explore integration techniques, real-world applications, and the benefits of enhanced interactivity. A hands-on demo with the Grok library demonstrates creating and connecting functions for financial tasks, equipping participants to build dynamic, task-driven AI applications."

This session introduces Streamlit, a Python library for rapid web app development, transitioning from theory to practical AI application-building. Participants recap key Generative AI topics, then explore Streamlit’s features like interactive widgets and real-time updates, ideal for AI projects. The session includes building a simple app, setting up a GitHub repository, and deploying it on Streamlit Cloud, providing hands-on experience and tools for future AI projects.

The Generative AI Program, participants learn to transition from Google Colab to Streamlit, focusing on code structure, token limits, and system message guardrails. They enhance the UI with widgets and CSS, create a GitHub repository for a Sachin Tendulkar chatbot, and deploy the app to Streamlit. The session concludes with testing the app's functionality and preparing for the next steps in deployment and chatbot limitations.

In this session, participants enhance the summarizer app by adding a PDF uploader, integrating multiple LLM models, and improving the UI with a dropdown for summary types. The session covers backend engineering for smooth model handling and concludes with deploying the app on Streamlit via GitHub.

The session focused on building a text summarizer app that condenses large texts into concise summaries. Topics included coding in Google Colab, Streamlit app building, securely storing API keys, configuring LLM outputs, dynamic prompt assignment, AI roles, guardrails, session state, and CI/CD. The app was tested in GitHub Codespace, deployed on Streamlit Cloud, and demonstrated secure API key storage and dynamic functionality.

In this session, participants enhance the summarizer app by adding a PDF uploader, integrating multiple LLM models, and improving the UI with a dropdown for summary types. The session covers backend engineering for smooth model handling and concludes with deploying the app on Streamlit via GitHub.

In this session, participants explore Hugging Face’s open-source tools, including the Model Hub, Datasets, and Spaces, to drive AI innovation. They learn about the Inference API for applications like chatbots and image generation, focusing on scalability and rate limiting. The session concludes with a hands-on demo, where participants set up accounts and deploy a Streamlit app for text-to-image generation.

In this session, participants explore image classification in generative AI, focusing on CNNs, data preprocessing, and model optimization techniques. They set up a GitHub project and develop a Streamlit app for gender classification and AI-driven image detection. The session concludes with deploying the app on Streamlit and discussing detection accuracy and threshold settings.

This session introduces "Fast HTML," a Python web framework known for speed, simplicity, and scalability. Participants explore its advantages over Streamlit, review SQLite integration, and learn deployment options. A hands-on demo covers creating applications with features like image generation and parallel prompting, concluding with running a Fast HTML app.

In this session, participants enhance their Fast HTML app with features like image deletion for storage management and an image counter to track usage. They implement a free image generation limit to encourage paid upgrades and prevent overuse. Hands-on demos cover feature implementation, setting limits, and managing upgrade prompts.

This session introduces Retrieval Augmented Generation (RAG) to enhance language model accuracy by grounding responses with retrieved knowledge. Participants learn about vector stores, embeddings, and their role in storing and retrieving unstructured data. The session covers tools like ChromaDB and Pinecone, highlighting their benefits and use cases.

This session dives into Retrieval Augmented Generation (RAG), focusing on reducing hallucination and grounding LLM responses. Participants learn to preprocess data, create embeddings, and store them in vector databases like Astra DB. A hands-on demo covers the complete RAG workflow, from document chunking to generating accurate, query-based responses.

This session delves into Retrieval Augmented Generation (RAG), covering data retrieval accuracy, multi-chunk fusion, and evaluation metrics like BLEU and ROUGE. Participants explore re-ranking, query expansion, and human feedback to refine RAG systems with hands-on demos and best practices.

In this session, participants build a local vector database from a PDF using Faiss, focusing on embedding storage and efficient similarity searches. They explore tools like Pickle files for secure data handling and caching for optimization, with hands-on demos showcasing dimensional embeddings and query retrieval accuracy.

Participants explore object detection in computer vision using YOLO, focusing on bounding boxes, confidence scores, and real-world applications. Hands-on exercises cover YOLO detection and LLM integration for summarizing detected objects and their details.

Participants explore image upscaling, enhancing resolution using AI techniques like GANs and VAEs compared to traditional methods. Hands-on exercises with Cloudinary demonstrate applications in photography, gaming, and medical imaging for sharper, realistic visuals.

In this session, participants explore generative AI fill to expand, repair, or modify images by generating seamless additional content. They learn techniques like texture synthesis, image outpainting, and background completion, with hands-on practice using Cloudinary to extend images for real-world applications in design and media.

In this session, participants explore Generative Replace, an AI technique for seamlessly substituting objects in images. They learn to use tools like Cloudinary to automate replacements, leveraging GANs and object detection for realistic results, with applications in e-commerce, marketing, and content creation.

In this session, participants explore Image-to-Text AI, which interprets images to generate descriptive text, enhancing accessibility and automating tasks like cataloging and storytelling. They learn about models like LAVA and LAMA 3.1, combining vision and language for detailed descriptions and narratives.

In this session, participants explore AI-driven image editing, enabling users to edit images through text commands like 'remove the chair,' without traditional skills. Using tools like Cloudinary’s Generative Remove, they learn object removal, non-destructive editing, and accessibility-focused techniques powered by advanced AI object detection.

In this session, participants explore AI-driven image recoloring, enabling quick, realistic color changes to specific items in images using text commands. Leveraging tools like Cloudinary’s API, they learn to efficiently showcase product variations with advanced object detection and natural language processing.

In this session, participants explore AI-powered image restoration, a fast and efficient technique to repair old or damaged photos using tools like Cloudinary. They learn about key technologies such as GANs, deep learning models, and image in-painting for seamless restoration, including a hands-on demonstration.