1 00:00:03,000 --> 00:00:06,219 Congratulations on completing the course. 2 00:00:06,219 --> 00:00:12,680 You have now gained a comprehensive understanding of building agentic AI applications with LangChain and LangGraph. 3 00:00:12,680 --> 00:00:17,579 You can start exploring opportunities to continue learning and applying your acquired skills. 4 00:00:17,579 --> 00:00:22,620 But before that, let's review some key concepts you've learned throughout the course. 5 00:00:22,620 --> 00:00:28,340 Generative AI is a reactive system that creates content like text or images based on prompts. 6 00:00:28,340 --> 00:00:31,239 It stops once it generates a response. 7 00:00:31,239 --> 00:00:35,580 Agentic AI, on the other hand, is proactive and uses prompts to pursue goals through a 8 00:00:35,580 --> 00:00:39,659 loop of decision-making, action, and learning. 9 00:00:39,659 --> 00:00:46,599 It can operate with minimal human input, making it far more dynamic and autonomous than traditional genAI. 10 00:00:46,599 --> 00:00:52,580 LangGraph structures AI workflows as flexible graphs using nodes, edges, and shared state. 11 00:00:52,580 --> 00:00:58,220 It supports looping, branching, state persistence, and human-in-the-loop functionality, making 12 00:00:58,220 --> 00:01:02,459 it ideal for building complex, adaptive agents. 13 00:01:02,459 --> 00:01:06,900 LangChain helps developers build LLM-powered applications using modular components like 14 00:01:06,900 --> 00:01:09,620 prompts, memory, and tools. 15 00:01:09,620 --> 00:01:13,019 It works best for linear, step-by-step workflows. 16 00:01:13,019 --> 00:01:19,040 LangGraph, on the other hand, extends LangChain's capabilities by enabling stateful, multi-agent 17 00:01:19,040 --> 00:01:22,260 workflows through graphs of nodes and edges. 18 00:01:22,260 --> 00:01:28,459 It's ideal for handling complex tasks that involve branching, memory, and long-term context. 19 00:01:28,459 --> 00:01:33,300 Here's a breakdown of how LangGraph structures and executes stateful workflows. 20 00:01:33,300 --> 00:01:36,739 TypedDict is used to define a structured state. 21 00:01:36,739 --> 00:01:43,620 State can include complex types like lists, nested dictionaries, or message sequences. 22 00:01:43,620 --> 00:01:48,099 Nodes are linked to functions that either transform or observe state. 23 00:01:48,099 --> 00:01:51,139 Edges define transitions between nodes. 24 00:01:51,139 --> 00:01:54,860 Conditional edges evaluate the current state to control flow. 25 00:01:54,860 --> 00:01:58,300 Functions use state unpacking to update state immutably. 26 00:01:58,300 --> 00:02:04,379 And finally, StateGraph is compiled and executed with an initial state using invoke. 27 00:02:04,379 --> 00:02:07,180 Here's a quick glance at reflection agents. 28 00:02:07,180 --> 00:02:13,779 Reflection agents are AI systems that iteratively improve outputs by analyzing and refining previous responses. 29 00:02:13,779 --> 00:02:19,179 The process involves two core LLM roles, a generator and a reflector. 30 00:02:19,179 --> 00:02:24,979 Basic reflection agents operate in feedback loops to refine results over multiple iterations. 31 00:02:24,979 --> 00:02:29,979 LangChain is used to set up structured prompts and memory for each agent role. 32 00:02:29,979 --> 00:02:34,699 LangGraph's message graph tracks message flow and defines agent state across turns 33 00:02:34,699 --> 00:02:38,179 using message types like human message and AI message. 34 00:02:38,179 --> 00:02:41,740 Next, here's a look at reflexion agents. 35 00:02:41,740 --> 00:02:46,940 Reflexion agents iteratively refine AI output using self-critique and external tools. 36 00:02:46,940 --> 00:02:54,020 They differ from basic reflection by integrating real-time data and generating structured, cited responses. 37 00:02:54,020 --> 00:02:59,539 Outputs follow a schema with labeled fields like response, critique, and citations. 38 00:02:59,539 --> 00:03:03,380 This loop continues until a verifiable result is achieved. 39 00:03:03,380 --> 00:03:08,960 Next up are ReAct agents, which reason iteratively and only use tools when necessary. 40 00:03:08,960 --> 00:03:15,399 Their structured response flow is Thought, Action, Action Input, Observation, and Final Answer. 41 00:03:15,399 --> 00:03:19,919 The agent maintains message history to inform each step of reasoning. 42 00:03:19,919 --> 00:03:24,440 Execution stops when no further tool calls are required, yielding the final answer. 43 00:03:24,440 --> 00:03:30,119 Here's a look at multi-agent systems, which consist of autonomous agents with specialized roles. 44 00:03:30,119 --> 00:03:34,600 They interact through graph-structured workflows using collaboration patterns like pipelines 45 00:03:34,600 --> 00:03:37,240 and hub-and-spoke models. 46 00:03:37,240 --> 00:03:44,320 Generalist agents often coordinate specialized agents, balancing depth of expertise with broad task management. 47 00:03:44,320 --> 00:03:50,300 Orchestration frameworks such as LangGraph, CrewAI, AutoGen, and BeeAI enable scalable and 48 00:03:50,300 --> 00:03:52,720 modular agent collaboration. 49 00:03:52,720 --> 00:03:59,119 Agentic RAG enhances RAG by letting an LLM act as a decision-making agent, not just a responder. 50 00:03:59,119 --> 00:04:03,139 The agents select the most relevant data source based on query context. 51 00:04:03,139 --> 00:04:08,539 It boosts accuracy, adaptability, and real-world applicability across industries. 52 00:04:08,539 --> 00:04:12,139 If you haven't yet enrolled in the Professional Certificate Program, of which this course 53 00:04:12,139 --> 00:04:14,899 is a part, we encourage you to do so. 54 00:04:14,899 --> 00:04:18,799 Depending on your schedule and the number of courses in the program, you can complete 55 00:04:18,799 --> 00:04:21,320 it in approximately two to six months. 56 00:04:21,320 --> 00:04:24,940 We recommend that you continue applying the knowledge gained from this course in your 57 00:04:24,940 --> 00:04:27,540 agentic AI development career. 58 00:04:27,540 --> 00:04:32,660 We hope these principles will refine your skills and empower you to advance professionally. 59 00:04:32,660 --> 00:04:35,019 Congratulations on completing this course! 60 00:04:35,019 --> 00:04:38,700 We appreciate your participation in this learning journey and wish you all the best!