Mastering Generative AI From Neural Networks to Multi-Agents

Learn the foundations of machine learning, including supervised, unsupervised, semi-supervised, and reinforcement learning, and master the end-to-end pipeline, data quality, feature engineering, and key evaluation metrics for generative ai.

Explore neural networks and deep learning as the foundation of generative AI, covering activation functions, backpropagation, training, and hands-on lab with architectures from feedforward to transformers and embeddings.

Explore embeddings as numeric vector representations that capture content similarity across text, images, and videos; learn training, vector space, similarity measures, and real-world applications like search, clustering, and recommendations.

Explore the transformer architecture, including self-attention, encoder–decoder structure, and autoregressive training, and learn how masking, positional encoding, and vocabulary shape modern language models.

Explore diffusion models that transform text prompts into images and videos, and learn their forward and reverse diffusion training, architecture, conditioning, seeds, and real-world applications.

Clone the GitHub project, open in VSCode, install UV, then install dependencies (Jupyter, AWS, torch, torchvision, transformers) and set up the Jupyter kernel to start with lab zero.

Design and train a PyTorch flower image classifier from scratch, handling data import, transforms, and forward passes, then evaluate with metrics and explore transfer learning with ResNet50.

Explore tokenization as the foundation for training large language models. Compare word-based, subword, and character approaches, and examine byte pair encoding, vocabulary size, and special tokens.

Turn unstructured documents into structured topics for search, tagging, and spam or sentiment detection. Compare classical machine learning, embeddings, and transformer models with preprocessing steps like tokenization, stemming, and tf-idf.

Master topic modeling to turn unstructured documents into embeddings, cluster them into coherent topics, assign keywords with tf-idf, and create crisp topic descriptions with an llm.

Explore named entity recognition to identify persons, organizations, and locations, and build a transformer-based solution with U/B/I/L labeling for single and multi-word, nested entities.

Explore pre-training of large language models through self-supervised objectives and a data preparation pipeline, yielding a base model later refined by supervised fine-tuning and reinforcement learning with human feedback.

Explore supervised fine-tuning of pre-trained foundation models to teach new skills and align behavior. Evaluate parameter-efficient methods—low rank adapters, adapters, prompt tuning, and prefix tuning—for task specialization and performance.

Explore how to evaluate large language models across translation, text generation, and search tasks, using metrics like bleu, rouge, meteor, bert score, and ranking benchmarks.

Explore how text-to-image and image-to-image diffusion models generate and modify images, using cross-attention conditioning, control nets, and adapters to guide style, content, and details.

Explore diffusion-based image editing with inpainting, outpainting, and mask-driven prompts, enhanced by control nets and segmentation masks.

Explore fine-tuning diffusion text-to-image models with Dreambooth, textual inversion, and laura to preserve subject details while reducing compute; train on few images with a unique identifier and prompts.

Set up an AWS account, enable MFA. Create a Bedrock access subuser, activate Bedrock models, and explore Nova models via the Bedrock playground while reviewing token-based pricing.

Explore image creation with Amazon Bedrock and Nova Canvas, using text prompts, masks, styles, and outpainting to generate and edit images via the api.

Master advanced image editing with Amazon Bedrock Nova models by using inpainting and outpainting, encoding images to base64, and invoking the Bedrock model API with prompts and masks.

Explore in-context learning and prompt engineering with pre-trained LLMs to teach new data, build chatbots and analytics pipelines, and generate reports using zero/few-shot learning, augmentation, and structured outputs.

Build a retrieval augmented generation (rag) pipeline that ingests documents, chunks content, creates embeddings, stores vectors in a vector database, and retrieves top-k chunks to augment prompts.

Discover GraphRAG and StructRAG to extend retrieval with graph and structured data, enabling LLMs to reason across entities, relationships, and tables.

Explore fine-tuning and alignment techniques for large language models, including supervised fine tuning, reinforcement learning from human feedback, and direct preference optimization, with practical guidance on when to use each.

Discover how model distillation compresses large language models into smaller student models, enabling fast, cost-efficient inference and deployment with accuracy close to the teacher, via supervised fine-tuning and reinforcement learning.

Explore embeddings with HuggingFace sentence transformer, visualize word and sentence vectors in two dimensions with t-SNE, compare search methods, and build retrieval augmented generation.

Explore vector databases with embeddings, FAISS, IVF, and ANN to enable exact and approximate searches and build a retrieval augmented generation movie recommender.

Explore Amazon Bedrock and prompt engineering with system and user prompts to define personas and tasks like sentiment analysis, summarization, and AI writing; compare models on latency and customization.

Explore multi-turn chat with system prompts, use Amazon Bedrock LLMs, and perform classification, summarization, and creative writing while comparing models and applying guardrails for robust ai applications.

learn to build a custom chatbot by combining neural networks, embeddings, and large language models with memory, chat history, and a knowledge base using rag techniques.

Build a local chatbot with memory and a Gradio UI using large language models and retrieval augmented generation over knowledge bases, featuring conversation history, embeddings, and a vector store.

Learn how agents extend large language models with planning, API discovery, and autonomous actions, using the react pattern and topologies like routing, sequential, swarm, mesh, and star.