1 00:00:03,000 --> 00:00:06,139 Congratulations on completing this course. 2 00:00:06,139 --> 00:00:10,180 You now have deep insight into the fundamentals of building AI agents. 3 00:00:10,180 --> 00:00:15,060 You're ready to apply your skills and pursue new opportunities to continue your learning. 4 00:00:15,060 --> 00:00:19,360 Before you move on, let's review your key takeaways from this course. 5 00:00:19,360 --> 00:00:25,319 You now know that tools help LLMs access external data and support RAG, enabling the use of 6 00:00:25,319 --> 00:00:28,959 the organization's or other specialized databases. 7 00:00:28,959 --> 00:00:34,880 Tools also help process images, audio, and video to enable vision, voice, and multimodal 8 00:00:34,880 --> 00:00:41,119 reasoning, manage long conversations, and connect to APIs to perform real-world actions. 9 00:00:41,119 --> 00:00:45,540 Embedded tool calling improves LLM accuracy and reduces hallucinations by centralizing 10 00:00:45,540 --> 00:00:50,319 tool handling within a dedicated library or framework, replacing error-prone client-side 11 00:00:50,319 --> 00:00:51,759 implementations. 12 00:00:51,759 --> 00:00:56,680 Tools help LLMs select the correct function by briefly describing their purpose, defining 13 00:00:56,680 --> 00:01:01,599 the expected output, and showing the required input and output formats. 14 00:01:01,599 --> 00:01:05,919 Agents select a step-by-step reasoning loop that selects a tool, calls the tool, reviews 15 00:01:05,919 --> 00:01:10,720 the result answer, feeds the result into the process, and repeats the process as needed 16 00:01:10,720 --> 00:01:14,480 until the LLM and tools produce a final response. 17 00:01:14,480 --> 00:01:19,580 The Zero-Shot React agent uses zero-shot reasoning to solve tasks it hasn't seen before and 18 00:01:19,580 --> 00:01:23,120 works best for simple or well-structured problems. 19 00:01:23,120 --> 00:01:27,599 When building an agent in LangChain, choose an LLM that supports tool use and complex 20 00:01:27,599 --> 00:01:33,040 reasoning, uses tools with JSON serializable inputs and outputs, and select an agent strategy 21 00:01:33,040 --> 00:01:35,480 that matches task complexity. 22 00:01:35,480 --> 00:01:40,160 LangChain lets you build flexible agents using create_react_agent to 23 00:01:40,160 --> 00:01:42,919 define how the agent thinks and responds. 24 00:01:42,919 --> 00:01:47,040 You can customize the agent's behavior by passing in your prompt templates and tool 25 00:01:47,040 --> 00:01:49,599 lists to control reasoning style. 26 00:01:49,599 --> 00:01:54,480 You can group tools such as addition, subtraction, multiplication, and division tools into a 27 00:01:54,480 --> 00:01:59,559 custom math toolkit, and the agent can use them to perform calculations, and use the 28 00:01:59,559 --> 00:02:06,440 .invoke method to simulate chat interactions, send messages to the agent, and receive structured 29 00:02:06,440 --> 00:02:07,440 responses. 30 00:02:07,440 --> 00:02:10,779 Next, review what you've learned about manual invocation. 31 00:02:10,779 --> 00:02:16,399 You learned that LLMs specify task parameters and suggest tools, and agents automatically 32 00:02:16,399 --> 00:02:19,520 invoke tools based on direction from the LLM. 33 00:02:19,520 --> 00:02:23,399 You know that manual invocation involves verifying inputs and outputs and adjusting 34 00:02:23,399 --> 00:02:28,899 actions as needed, and that manual invocation provides greater control that can help organizations 35 00:02:28,899 --> 00:02:33,399 enhance safety, manage costs, and deliver more accurate information. 36 00:02:33,399 --> 00:02:37,600 You've learned that you can analyze and visualize data by asking natural language 37 00:02:37,600 --> 00:02:40,320 questions to the Pandas DataFrame agent. 38 00:02:40,320 --> 00:02:45,240 The Pandas DataFrame agent generates Python code that directly interacts with your DataFrame, 39 00:02:45,240 --> 00:02:50,360 filtering, aggregating, and visualizing data based on your natural language prompts. 40 00:02:50,360 --> 00:02:55,360 Your DataFrame refers to the Pandas DataFrame object that you loaded or created and then 41 00:02:55,360 --> 00:02:57,880 passed to the LangChain Pandas agent. 42 00:02:57,880 --> 00:03:02,880 You also know that LCEL pattern structures workflows use the pipe operator for precise 43 00:03:02,880 --> 00:03:04,399 data flow. 44 00:03:04,399 --> 00:03:09,240 You can define prompts using templates that include variables and side {}. 45 00:03:09,240 --> 00:03:14,279 You can link components for sequential execution using runnable sequence, and run multiple 46 00:03:14,279 --> 00:03:18,600 components concurrently on the same input using runnable parallel. 47 00:03:18,600 --> 00:03:24,119 You simplify syntax by replacing runnable sequence with the pipe operator, and LCEL 48 00:03:24,119 --> 00:03:29,679 automatically converts functions and dictionaries into compatible components using type coercion. 49 00:03:29,679 --> 00:03:34,960 No review would be complete without examining best practices when beginning to use agents. 50 00:03:34,960 --> 00:03:40,279 Always start with sandboxed environments, design clear prompts, validate the LLM analysis 51 00:03:40,279 --> 00:03:45,639 with human expertise, and iteratively refine your queries for safe and effective AI-driven 52 00:03:45,639 --> 00:03:46,639 analysis. 53 00:03:46,639 --> 00:03:52,479 AI-powered SQL agents allow a broader range of users to access and interpret data without 54 00:03:52,479 --> 00:03:54,679 needing deep technical skills. 55 00:03:54,679 --> 00:03:59,039 To develop applications that use natural language queries with LangChain, start by creating 56 00:03:59,039 --> 00:04:01,259 a Python virtual environment. 57 00:04:01,259 --> 00:04:06,440 Then install the required libraries, such as LangChain and your LLM, launch the SQL 58 00:04:06,440 --> 00:04:09,080 server, and build a database connector. 59 00:04:09,080 --> 00:04:13,399 Once your environment is ready, you can run natural language queries using LangChain's 60 00:04:13,399 --> 00:04:19,160 SQL agent, which translates those queries into SQL and retrieves results from the database. 61 00:04:19,160 --> 00:04:22,839 With this knowledge, we encourage you to continue learning and apply this course towards an 62 00:04:22,839 --> 00:04:25,239 IBM Professional Certificate. 63 00:04:25,239 --> 00:04:28,760 Depending on your schedule and the number of courses in the program, you can complete 64 00:04:28,760 --> 00:04:32,200 this Professional Certificate in approximately two months. 65 00:04:32,200 --> 00:04:35,679 You'll find the links to the Professional Certificate and several related courses in 66 00:04:35,679 --> 00:04:39,500 the congratulations and next steps reading at the end of this course. 67 00:04:39,500 --> 00:04:42,399 We encourage you to continue practicing your new skills. 68 00:04:42,399 --> 00:04:47,040 We hope your new skills support your current work and enable you to advance your career. 69 00:04:47,040 --> 00:04:49,320 Congratulations on completing this course. 70 00:04:49,320 --> 00:04:52,679 We appreciate your participation in this learning journey and wish you all the best.