week 6

Capstone

Wrap up with final team projects: build a usable AI agent MVP for a specific use case (with low-code tools). Review recently released tools: e.g. OpenAI’s Assistants API (a stateful, tool-enabled assistant interface) and consider how it could fit into architectures. Discuss future trends (ChatGPT plugins, GPT-4o, etc.) and continuous learning cycles.

• End-to-end system design, demos.
• Projects:
  
  • Build a customer support system with:
    
    • Multi-agent collaboration
    • RAG-powered knowledge base
    • Deployment (Netlify + AWS/Azure).

Demo Day: Present projects to peers/instructors.

week 5

Deploy, Best Practices

Day 9

• Database integration (SQL, NoSQL, vector stores)
• Deployment options: serverless, containers, edge
• Monitoring and observability for AI agents
• Hands-on: Deploying an agent to a cloud platform with proper monitoring

Day 10

• Best Practices and Product Integration
  
  • User experience considerations
  • Cover prompt and agent best practices: clear instructions, safety checks, fallback strategies.
  • Emphasize human-in-the-loop controls (e.g., having a human review critical agent actions.
  • Discuss performance monitoring (e.g,. using LangSmith or logs).
  • Address AI limitations: biases, hallucinations, and how RAG/human oversight mitigates them.
• Finetuning basics (out of scope for this. Too technical. Depends on the pace & interest of the class)

week 4

MCPs, Optimize

Day 7: Model Context Protocol

• Understanding context windows and their limitations
• Context management strategies
• Model Context Protocol principles and implementation
• Hands-on: Implementing effective context management in agent design

Day 8: Caching and Optimization Strategies

• Semantic caching principles
• Dynamic vs. static caching strategies
• Optimizing for latency, costs, and performance
• Hands-on: Implementing caching systems for agent responses
• Token economics and cost structures

week 3

Agentic Architecture, Tooling, Guardrails

Day 5: Agentic design patterns and roles. Multi-agent architectures (routers, planners, specialists). Frameworks like Microsoft’s AutoGen for coordinating multiple agents. Using single-agent vs multi-agent designs (e.g. a triage agent dispatching tasks to specialized agents).

• Types of AI agents: simple reflex, model-based, goal-based, utility-based
• Agent architectures: reactive, deliberative, hybrid
• Planning frameworks for agent design
• User journey mapping for agent interactions
• Designing agent teams with specialized roles
• Communication protocols between agents
• Conflict resolution and collaboration patterns
• Hands-on: Creating a simple multi-agent system

Day 6: Focus on building working prototypes with minimal code. Review no-code/low-code platforms for agents (LangFlow, n8n, Zapier with OpenAI plugins, etc.). Discuss integrating agents into product interfaces (chat UIs, Slack bots, webhooks).

• Function calling and API integration
• Authentication and security considerations
• Rate limiting and error handling
• Guardrails: Avoiding hallucinations, bias, and harmful outputs.
• Hands-on: Connecting an agent to external APIs

week 2

Getting Started with Agents, RAG

Day 3: Vector databases and memory buffers for stateful agents.

• Understanding embeddings and vector representations
• Understanding vector embeddings in depth
• Short-term vs. long-term memory patterns
• Vector database options (Pinecone, Chroma)
• Chunking strategies and metadata management (may be not)
• Hands-on: Setting up a vector database and optimizing retrieval quality

Hands-on: Experimenting with embeddings and similarity matching

‍

Day 4: How agents use external knowledge: Retrieval-Augmented Generation (RAG). Why grounding on documents or data sources improves factuality and relevance.

• Build a drag-and-drop agent in 15 minutes
• RAG architecture and components
• Preparing data for agents (e.g., formatting documents for RAG)
• Popular RAG frameworks (LlamaIndex, LangChain)
• Evaluation metrics for RAG systems
• Hands-on: Building a simple RAG system with your own documents
• Play with Notebooklm

‍

week 1

Starting with Why + Foundations of LLMs and AI Agents (Workflows), Prompt Engineering

Day 1: Getting Started, Journey to AI Agents, Basics of LLMs, When and why go agentic

• What and Why of AI and Agents?
• Start with design - Case study to visualize the Workflow: Create an end-to-end workflow of an existing complex system. Explore examples of agentic systems and use cases (e.g., customer support bots, scheduling assistants).
• Understand the evolution of ML, AI, and Gen AI via the case study (LSTM, Attention is all you need, other?)
• Introduction to LLMs and the notion of an “AI agent. Discuss how AI agents differ from today’s software, simple chatbots, or fixed pipelines.

‍

Day 2: Prompt Engineering - Basics to Advanced

• Prompt Engineering (Zero Shot, One Shot, Few Shot, Chain of thought, ReAct, Prompt Injection, Role-based, Tree of thoughts, Automatic, Meta, Contextual, etc.)
• Hands-on: Exploring existing AI agent platforms (ChatGPT, No code UI development - Lovable, v0, Gamma, Napkin)

‍