Course Details

Courses Descriptions

MODULE OBJECTIVE

What you'll achieve

Build AI agents from first principles, design stateful LangGraph workflows, implement all major agent patterns, compare CrewAI vs AutoGen, and ship a production multi-agent e-commerce support system.

SECTION 01 — LangChain in Depth

Foundation models, tools, memory, MCP

• Prompt vs Fine-Tuning vs RAG vs Agentic decision map
• Get Ready: API, Structured Output (using Schema), Custom Tool: Search Tools (Agent), Memory, State, Multimodal, Project: (Personal Chef)
• Get Application Ready: Runtime, MCP (intro), Sub Agent, Multi Agent, RAG-SQL Agent, Project: (Wedding Planner)
• Get Production Ready: Message, Memory, Dynamic: Models, Prompts, Tools, Human in the loop, Project: (Email Agent)
• Utilizing MCP Servers

SECTION 02 — LangGraph in Depth

Stateful graph-based agent workflows

• Python: State Variable, Node definitions, Pydantic
• LangGraph: StateGraph, Nodes, Edges, Shared State, Graph Builder and Graph Visualizing
• Sequential Workflow, Multiple Inputs, Conditional Routing, Looping, React
• LKTM Framework: L: LLM, M: Memory, Short Term, Long term, Working memory, K: Knowledge Base, T: Local, API, MCP, Pydantic, Reducer Logic, Schema Modification

SECTION 03 — Multi-User Chat Memory

Isolated per-user conversational state

• Thread-based conversation isolation
• Token-controlled memory with context trimming
• Persistent InMemorySaver checkpointer
• Switching between users and memory view

SECTION 04 — ReAct vs ReWOO

Reasoning architecture deep dive

• ReAct: Reason + Act interleaved loop
• ReWOO: Reasoning Without Observation (plan-first)
• Head-to-head: research assistant task
• Decision framework + hybrid approaches

SECTION 05 — Reflection Agents

Self-correcting code generation

• Generator → Evaluator → Reflector loop
• Code Writer + Code Reviewer agent pair
• Iterative self-correction until tests pass

SECTION 06 — Parallel Execution & Send

Fan-out, map-reduce, dynamic parallelism

• Linear vs parallel workflow comparison
• Annotated state with operator.add for fan-out
• Send construct: dynamic parallel execution
• Map-Reduce for scalable batch processing

SECTION 07 — Agentic RAG

RAG design patterns in LangGraph

• Router RAG, Corrective RAG (CRAG)
• Self-Reflective RAG, Adaptive RAG
• Speculative RAG, Self-Routing RAG
• Complexity vs flexibility comparison guide

SECTION 08 — Reflective Planning Agent

Multi-step complex query project

• Reflective Dynamic Planning Agent architecture
• Web research sub-agent with tool-use graph
• Plan → Execute → Replanner → reflection loop
• End-to-end: multi-step complex query analysis

SECTION 09 — HITL · Supervisor · Interrupts

Safety and human oversight

• HITL: reviewing tool calls, validating outputs, Interrupt
• construct, Command class: approve/reject Subgraphs and supervisor architecture, Multiturn Conversion
• FailSafe patterns for production systems

SECTION 10 — CrewAI & AutoGen Frameworks

Multi-framework comparison

• CrewAI: role-based agents, hierarchical mode
• AutoGen: GroupChat, nested chat, sequence agents
• LangGraph Supervisor vs CrewAI vs AutoGen
• Framework selection decision guide

SECTION 11 — Design Pattern & Evaluation

Coordination pattern mastery

• Design Patterns using Langgraph: Single Agent Pattern, ReAct Pattern, Custom Logic Pattern, Sequential Pattern, Parallel Pattern, Loop Pattern
• Quality and Refinement Pattern, Iterative Refinement Pattern, Review & Critique Pattern, Orchestration & Emergent Pattern
• HITL, Trust & Reputation, Governance
• Evaluating Agents using Langsmith, DeepEval, Arize
• Deployment, Monitoring and QC of Agents using Langsmith

SECTION 12 — Capstone: E-Commerce Support

Production multi-agent build

• ShopSmart multi-agent customer support architecture
• PII redaction, RAG knowledge base, Supervisor router
• Specialist agents: Orders, Returns, Billing, Product
• HITL escalation for Platinum customers
• Login-based e-commerce deployment + system metrics

KEY TAKEAWAYS

1. Agent = LLM + tools + memory + goal

2. ReAct is foundational — master it before multi-agent systems

3. Agentic RAG combines retrieval intelligence with agent reasoning

4. Understanding why 40–80% of naive MAS implementations fail

5. CrewAI = role-based; AutoGen = conversation-based flexibility

6. LangSmith tracing separates a demo from a production system