Applied Deep Learning with PyTorch

• Introduction to PyTorch: tensors, automatic differentiation, and the computation graph

• PyTorch vs other frameworks: why researchers and developers prefer it

• Building neural networks using torch.nn and nn.Module

• Implementing forward passes and activation functions

• Loss functions and optimizers using torch.optim

• Writing custom training and validation loops

• Efficient data loading using torch.utils.data.Dataset and DataLoader

• Handling overfitting with dropout, regularization, and early stopping

• Visualizing metrics and loss curves with TensorBoard

• Saving and loading models using state_dict and checkpoints

• Introduction to Intel Arc GPU architecture and AI acceleration

• Setting up PyTorch with Intel® Extension for PyTorch (IPEX)

• Training models on Intel Arc GPU

• Case study: Training an image classifier on Intel Arc GPU

• Best practices for deploying models trained with PyTorch

Introduction to Computer Vision with CNNs

• Introduction to Computer Vision and real-world applications

• Challenges in interpreting images through AI

• Image representation: pixels, channels, resolution, and color spaces

• Why CNNs? Understanding the limitations of traditional neural networks for vision tasks

• Fundamentals of Convolutional Neural Networks (CNNs)

• Building a simple CNN architecture from scratch

• Image classification workflow using TensorFlow/Keras or PyTorch

• Working with image datasets (CIFAR-10, MNIST, etc.)

• Data preprocessing and augmentation techniques

• Model training, validation, and evaluation

• Visualization of learned filters and feature maps

• Transfer learning and using pre-trained CNNs (VGG, ResNet, MobileNet)

• Performance tuning and avoiding overfitting in CNNs

• Hands-on: Building a digit/image classifier using CNN

• Introduction to edge deployment on AI-enabled devices

Natural Language Processing (NLP)

• Introduction to NLP and its real-world applications

• Understanding structured vs unstructured text data

• Text preprocessing techniques

• Working with libraries like NLTK and spaCy
• Text representation methods
• Word embeddings
• Text classification basics
• Sequence models overview
• Hands-on: Building a sentiment analysis classifier
• Evaluating NLP models: accuracy, precision, recall, F1-score
• Introduction to transfer learning in NLP
• Ethical challenges in NLP: bias and misinformation

Introduction to Generative AI

• What is Generative AI?

• Real-world applications of generative models

• Generative vs traditional AI: key differences

• Overview of foundational models

• Understanding autoencoders and latent spaces

• Introduction to Generative Adversarial Networks (GANs)

• Introduction to Diffusion Models

• Introduction to Large Language Models (LLMs)

• Creative tools powered by Gen AI

• Ethical considerations in content generation

• Understanding hallucinations and content validation

• The role of prompt engineering in content control

• Hands-on demo

• Introduction to open-source generative AI tools

• The future of Generative AI and emerging trends

Large Language Models (LLMs) and Prompt Engineering

• Introduction to LLMs: what they are and how they work

• Overview of popular models: GPT, LLaMA,Claude, Mistral, and Gemini

• LLM capabilities and limitations

• Tokenization and model context window

• Few-shot, zero-shot, and chain-of-thought prompting

• Prompt Engineering fundamentals

• Prompt patterns for specific tasks

• Using OpenAI’s GPT and HuggingFace models via APIs

• Tools for prompt testing and refinement

• Introduction to prompt chaining and dynamic prompt templates

• Safety, alignment, and ethical concerns when working with LLMs

• Hands-on: Crafting and refining prompts for real-world tasks

Building AI-Powered Applications with Flask and Streamlit 

• Introduction to web applications for AI and ML

• Overview: Flask vs Streamlit – use cases and strengths

• Basics of Flask

• Basics of Streamlit

• Structuring a basic AI application

• Integrating pre-trained ML/LLM models into apps

• Connecting front-end inputs to back-end predictions

• Uploading files and handling user input

• Hosting local vs cloud-based applications

• Debugging and testing application workflows

• Preparing apps for deployment (Heroku, Render, or cloud)

• UX tips for building user-friendly AI interfaces

Advanced Prompt Engineering and LLM Fine-Tuning

• Beyond the basics: Prompt refinement strategies for precision

• System prompts and role-based persona engineering

• Few-shot learning with structured examples

• Chain-of-thought prompting and multi-step reasoning

• Embedding-based prompts and semantic similarity

• Prompt chaining for multi-turn, multi-step tasks

• Retrieval-augmented prompting vs vanilla prompting

• Controlling tone, length, style, and response format

• Function calling and tool use with LLMs (OpenAI, Claude)

• Limitations of prompting and when fine-tuning is needed

• Introduction to fine-tuning LLMs

• Datasets for fine-tuning (custom, open-source)

• Using Hugging Face tools to fine-tune a model (e.g., LLaMA, Mistral)

• Evaluating fine-tuned models: performance and generalization

• Best practices for safe and ethical deployment

Reinforcement Learning: Fine-tune generative models through feedback-driven learning

• Overview of RL in the context of Generative AI

• What is Reinforcement Learning from Human Feedback (RLHF)?

• Role of feedback loops in fine-tuning LLM behavior

• Architectures that support RLHF (e.g., PPO, DPO)

• Stages of RLHF

• Algorithms used in RLHF

• Designing reward models for generative tasks

• Aligning language models with ethical guidelines and safety constraints

• Case study: How OpenAI uses RLHF in ChatGPT

• RLHF vs supervised fine-tuning – when to use what

• Challenges in scaling RL for large models

• Bias, interpretability, and feedback reliability

• Hands-on example: Use a small-scale transformer with simulated feedback

• Tools and libraries: TRL (Hugging Face), Accelerate, RLlib

Retrieval-Augmented Generation (RAG) for AI Models

• What is Retrieval-Augmented Generation (RAG) and why it matters

• Key components of a RAG pipeline

• Vector databases: FAISS, Pinecone, Weaviate overview

• Document chunking strategies for optimal retrieval

• Text embeddings and semantic search

• Creating knowledge bases from PDFs, HTML, Notion, etc.

• Connecting LLMs with retrieval layers using LangChain

• Prompting with contextual information from retrievers

• RAG vs traditional search vs fine-tuning: when to use what

• Deploying a basic RAG-powered Q&A app

• Evaluation metrics: relevancy, latency, hallucination reduction

• Best practices for scaling RAG systems

• Privacy and compliance in RAG-powered enterprise AI

• Case studies: RAG in customer support, internal documentation bots

Building a Local Retrieval-Augmented Generation (RAG) System on Intel AI PC

• Setting Up the Intel AI PC Environment
• Developing the RAG Pipeline
• Optimization Techniques for Intel Hardware
• Testing and Evaluation of the RAG System
• Deployment and Maintenance
• Hands on: Building a local RAG based application

Agentic AI – Autonomous AI Systems

• What is Agentic AI and how it differs from traditional AI models

• Use cases of AI agents in research, productivity, business, and automation

• Core components of an AI agent: memory, planning, reasoning, and tool use

• Frameworks for building agents

• Agent architecture and task decomposition

• Integrating tools and APIs into agent workflows

• ReAct (Reasoning + Acting) and other planning strategies

• Short-term vs long-term memory in agent systems

• Using vector databases and document loaders as knowledge tools

• Creating multi-agent systems with specialized roles

• Orchestration of concurrent agents for complex goals

• Logging, monitoring, and controlling autonomous behavior

• Security, constraints, and ethical boundaries in autonomous AI

• Hands-on: Build a research assistant or data analyst AI agent

• Future directions: self-improving and recursive agents

Setting up Docker for AI Development

• Introduction to Docker
• Installing and Configuring Docker
• Creating a Containerized AI Development Environment
• Running and Managing ContainersRunning and Managing Containers
• Docker Compose for Multi-Service AI Projects
• Best Practices

Cloud Deployment for LLM-Based Applications

• Introduction to cloud computing: key concepts and benefits

• Overview of major platforms: AWS, Azure, GCP, and Intel Developer Cloud

• Cloud infrastructure essentials: compute, storage, networking

• Hosting and deploying containerized AI apps with Docker

• Using cloud services to run FastAPI or Streamlit apps

• Deploying LLM APIs (e.g., OpenAI, Hugging Face) via cloud functions or REST endpoints

• Environment setup and configuration

• Introduction to serverless computing (Lambda, Azure Functions, Google Cloud Functions)

• CI/CD pipelines for AI app updates

• Working with secrets, environment variables, and model endpoints securely

• Logging, monitoring, and scaling applications in the cloud

• Using GPU instances to run inference with large models

• Cost management and optimization tips

• Hands-on: Deploy a simple LLM-powered chatbot or document Q&A system on the cloud

• Bonus: Introduction to Intel Developer Cloud for AI acceleration