Generative AI (Agentic AI)

Module 1: Foundations of AI, Machine Learning & Python for AI

• Introduction to AI & ML:
• What is AI? History and evolution
• AI vs. ML vs. Deep Learning vs. Generative AI vs. Agentic AI.
• Types of AI: Supervised, Unsupervised, Reinforcement Learning.
• Applications of AI across industries.

• Python Programming for AI:
• Python basics: data types, control flow, functions, OOP.
• Essential libraries: NumPy (numerical computing), Pandas (data manipulation), Matplotlib/Seaborn (data visualization).
• File handling and data loading.

• Machine Learning Fundamentals:
• Linear Regression, Logistic Regression, Decision Trees, K-Nearest Neighbors.
• Model evaluation metrics (accuracy, precision, recall, F1-score).
• Introduction to Scikit-learn.

• Deep Learning Fundamentals:
• Neural Networks: Perceptrons, Multi-Layer Perceptrons (MLPs).
• Activation functions, loss functions, optimizers (Gradient Descent, Adam).
• Backpropagation.
• Introduction to TensorFlow/Keras or PyTorch.

Module 2: Generative AI: Models and Techniques

• Introduction to Generative AI:
• What is Generative AI? Core concepts and types of generative models.
• Applications of Generative AI (text, image, video, audio generation).

• Generative Adversarial Networks (GANs):
• Architecture: Generator and Discriminator.
• Training GANs, common challenges (mode collapse, training instability).
• Types of GANs: DCGAN, Conditional GANs, CycleGAN, StyleGAN.
• Hands-on: Implementing a simple GAN for image generation.

• Variational Autoencoders (VAEs):
• Architecture: Encoder and Decoder.
• Latent space representation, reconstruction loss, KL divergence.
• Applications: data compression, anomaly detection, data generation.
• Hands-on: Building and training a VAE.

• Transformer Architecture:
• Self-attention mechanism.
• Encoder-Decoder architecture.
• Positional Encoding.
• Evolution from RNNs/LSTMs.

• Large Language Models (LLMs):
• Introduction to LLMs: GPT, BERT, LLaMA, Gemini.
• Pre-training and fine-tuning.
• Tokenization and embeddings.
• LLM APIs (OpenAI API, Google Gemini API, Hugging Face).

• Prompt Engineering:
• Fundamentals of prompt engineering: clear, concise prompts.
• Advanced techniques: Few-shot prompting, Chain-of-Thought (CoT), Tree-of-Thought (ToT), Self-reflection.
• Prompt optimization for specific tasks.
• Hands-on: Experimenting with various prompt engineering strategies.

• Retrieval-Augmented Generation (RAG):
• Limitations of standalone LLMs.
• Architecture of RAG: Retrieval system + Generative model.
• Vector databases (ChromaDB, Pinecone, FAISS): indexing, embeddings, similarity search.
• Implementing RAG for question answering and knowledge retrieval.
• Hands-on: Building a RAG system for a specific domain.

• Fine-tuning Generative Models:
• Why fine-tune? Transfer learning.
• Techniques: Full fine-tuning, LoRA (Low-Rank Adaptation), QLoRA.
• Dataset preparation for fine-tuning.
• Hands-on: Fine-tuning a pre-trained LLM for a custom dataset.

• Generative AI for Image and Other Modalities:
• Diffusion Models (Stable Diffusion, DALL-E): principles and applications.
• Text-to-Image generation: understanding prompts and parameters.
• Introduction to other generative models (music, video, code generation).
• Hands-on: Generating images with diffusion models.

Module 3: Agentic AI: Architecture and Development

• Introduction to Agentic AI:
• What are AI Agents? Definition, characteristics, and evolution.
• Distinction from traditional AI and Generative AI.
• Components of an intelligent agent: Perception, Cognition/Planning, Action, Memory.
• Types of agents: Reflex agents, Goal-based agents, Utility-based agents, Learning agents.

• Agent Architectures and Design Patterns:
• The Agentic Loop: Observe-Think-Act.
• Planning and Reasoning: Task decomposition, sub-goal generation.
• Tool Use: Enabling LLMs to interact with external tools and APIs.
• ReAct (Reasoning and Acting) and ReWOO (Reasoning with Open Ontology) patterns.

• Agentic Frameworks and Libraries:
• LangChain:
  • Chains and Agents.
  • Memory management.
  • Tool integration and custom tools.
  • Prompt templates.
  • LangChain Expression Language (LCEL).
  • Hands-on: Building basic agents with LangChain.
• LangGraph:
  • Building stateful, multi-step agentic workflows.
  • Graph-based agent design.
  • Human-in-the-Loop systems.
  • Hands-on: Developing complex agent workflows with LangGraph.
• AutoGen:
  • Multi-agent conversation frameworks.
  • Role-playing and collaborative agents.
  • Human-in-the-Loop with AutoGen.
  • Hands-on: Creating a multi-agent system for a specific task (e.g., research team).
• CrewAI (Optional/Advanced):
  • Orchestrating multi-agent teams.
  • Defining roles, tasks, and processes for collaborative AI.
  • Hands-on: Building a team of AI agents for a business process.
• Memory and State Management for Agents:
• Short-term vs. Long-term memory.
• Episodic memory, declarative memory.
• Integrating vector databases for external knowledge.
• Managing conversational context across turns.
• Building Tool-Using Agents:
• Designing and implementing custom tools for agents.
• Function calling in LLMs.
• Integrating with external APIs (web search, calendar, CRM, databases).
• Hands-on: Building agents that interact with external services.
• Multi-Agent Systems:
• Coordination and communication between agents.
• Agent collaboration patterns.
• Decentralized vs. centralized agent architectures.
• Use cases for multi-agent systems (e.g., supply chain optimization, automated customer support).
• Agentic RAG:
• Advanced RAG architectures using agentic principles.
• Agents for intelligent document retrieval and processing.
• Combining reasoning with retrieval for more accurate responses.

Module 4: Deployment, Evaluation, and Ethical AI

• Deployment Strategies for Generative AI and Agents:
• Containerization (Docker).
• Cloud deployment platforms (AWS SageMaker, Google Cloud AI Platform, Azure ML).
• API development (FastAPI, Flask) for AI services.
• Serverless functions for scalable deployment.
• Evaluation of Generative AI Models:
• Metrics for text generation (BLEU, ROUGE, perplexity).
• Metrics for image generation (FID, Inception Score).
• Human evaluation and qualitative assessment.
• Evaluation and Monitoring of AI Agents:
• Task success rate.
• Efficiency metrics (token usage, latency).
• Robustness and error handling.
• Observability tools (LangSmith, Langfuse).
• AIOps for agent monitoring and maintenance.
• Ethical AI and Responsible AI Development:
• Bias and fairness in Generative AI outputs and Agent decisions.
• Hallucinations and misinformation.
• Privacy concerns (PII, data leakage).
• Copyright and intellectual property issues in generated content.
• Security vulnerabilities in AI systems.
• Principles of Responsible AI (transparency, accountability, safety).
• Implementing guardrails and safety mechanisms.
• Future Trends in Generative and Agentic AI:
• Hybrid AI systems.
• Self-evolving agents.
• Multi-modal agents (integrating text, vision, speech).
• AI for scientific discovery and research.
• The evolving role of human-AI collaboration.

Module 5: Capstone Project

• Project Proposal & Design
• Students propose a real-world problem or application leveraging both Generative AI and Agentic AI.
• Define project scope, architecture, and required tools.
• Development & Implementation:
• Implement the proposed solution, applying learned concepts and frameworks.
• Iterative development and testing.
• Evaluation & Refinement:
• Rigorously evaluate the project's performance, safety, and effectiveness.
• Refine the models and agents based on evaluation results.
• Presentation & Documentation:
• Present the final project, showcasing functionalities and insights.
• Submit detailed technical documentation.
Book Now