1 00:00:04,000 --> 00:00:10,000 Hello, and welcome to this video on the LangChain LCEL Chaining Method. 2 00:00:10,000 --> 00:00:12,800 After watching this video, you'll be able to 3 00:00:12,800 --> 00:00:19,100 Describe how to build flexible, composable chains using LangChain's modern approach to prompt engineering. 4 00:00:19,100 --> 00:00:22,079 Structure prompts effectively using templates. 5 00:00:22,079 --> 00:00:25,840 Connect components using the pipe operator to streamline workflows. 6 00:00:25,840 --> 00:00:29,959 And develop reusable patterns for a variety of AI applications. 7 00:00:29,959 --> 00:00:34,860 LangChain Expression Language (or LCEL) is a pattern for building LangChain applications 8 00:00:34,860 --> 00:00:38,080 that utilizes the pipe (|) operator to connect components. 9 00:00:38,080 --> 00:00:41,559 This approach ensures a clean, readable flow of data from input to output. 10 00:00:41,559 --> 00:00:46,119 LangChain has evolved significantly, and this video will focus on the newer, recommended 11 00:00:46,119 --> 00:00:50,479 LCEL pattern rather than the traditional LLM chain approach. 12 00:00:50,479 --> 00:00:55,759 This modern method provides better composability, clearer visualization of data flow, and greater 13 00:00:55,759 --> 00:00:58,560 flexibility when constructing complex chains. 14 00:00:58,560 --> 00:01:02,000 To create a typical LCEL pattern, you need to 15 00:01:02,000 --> 00:01:05,040 Define a template with variables and curly braces 16 00:01:05,040 --> 00:01:12,220 Create a prompt template instance. Build a chain using the pipe operator to connect components. Invoke the chain with input values. 17 00:01:12,220 --> 00:01:15,919 Let's see this in action with a concrete example. 18 00:01:15,919 --> 00:01:19,839 In LangChain, runnables serve as an interface and building blocks that connect different 19 00:01:19,839 --> 00:01:24,680 components like LLMs, retrievers, and tools into a pipeline. 20 00:01:24,680 --> 00:01:27,860 There are two main runnable composition primitives. 21 00:01:27,860 --> 00:01:34,919 Runnable sequence chains components sequentially, passing the output from one component as input to the next. 22 00:01:34,919 --> 00:01:39,699 RunnableParallel runs multiple components concurrently while using the same input for each. 23 00:01:39,699 --> 00:01:43,779 However, LCEL provides elegant syntax shortcuts. 24 00:01:43,779 --> 00:01:48,260 For example, instead of using runnable sequence, the same sequential chain can be created by 25 00:01:48,260 --> 00:01:54,099 simply connecting runnable 1 and runnable 2 with a pipe, making the structure more readable and intuitive. 26 00:01:54,099 --> 00:01:58,099 LCEL also handles type coercion automatically. 27 00:01:58,099 --> 00:02:01,339 This means it converts regular code into runnable components. 28 00:02:01,339 --> 00:02:06,819 When you use a dictionary, it becomes a runnable parallel, which runs multiple tasks simultaneously. 29 00:02:06,819 --> 00:02:11,460 When you use a function, it becomes a RunnableLambda, which transforms inputs. 30 00:02:11,460 --> 00:02:15,759 This happens behind the scenes, so you don't have to handle the conversion manually. 31 00:02:15,759 --> 00:02:20,660 For example, in this code, the pipe operator combines the prompt templates with the LLM. 32 00:02:20,660 --> 00:02:26,619 The dictionary structure creates a RunnableParallel which processes all three tasks simultaneously. 33 00:02:26,619 --> 00:02:31,179 Each task receives the same input, text, but processes it differently. 34 00:02:31,179 --> 00:02:34,479 When you run this, it automatically becomes a RunnableParallel. 35 00:02:34,479 --> 00:02:41,259 The result will contain three keys, summary, translation, and sentiment, each with the output from the respective LLM call. 36 00:02:41,259 --> 00:02:45,039 Let's see LCEL in action by creating a simple chain. 37 00:02:45,039 --> 00:02:49,020 This code demonstrates how components can be connected using the pipe operator. 38 00:02:49,020 --> 00:02:53,259 The RunnableLambda in this chain wraps the format_prompt function, transforming it into 39 00:02:53,259 --> 00:02:55,860 a runnable component that LangChain can work with. 40 00:02:55,860 --> 00:03:00,220 When the chain runs, RunnableLambda takes the input dictionary, containing adjective 41 00:03:00,220 --> 00:03:04,059 and content keys, passes this dictionary to the format_prompt function. 42 00:03:04,059 --> 00:03:07,059 The function formats the prompt template with these variables. 43 00:03:07,059 --> 00:03:10,699 The formatted prompt is then passed to the next component, the LLM. 44 00:03:10,699 --> 00:03:14,660 The pipe operator creates a sequence by connecting runnable components together. 45 00:03:14,660 --> 00:03:19,419 In this joke chain, first, the RunnableLambda formats the prompt with variables. 46 00:03:19,419 --> 00:03:23,339 The pipe operator passes the formatted prompt to the LLM. 47 00:03:23,339 --> 00:03:27,460 Another pipe passes the LLM's response to the StrOutputParser. 48 00:03:27,460 --> 00:03:32,899 We've covered the essentials of LCEL, from its benefits and composition primitives to 49 00:03:32,899 --> 00:03:36,020 building practical chains using the pipe operator. 50 00:03:36,020 --> 00:03:40,339 Keep in mind that LCEL is best suited for simpler orchestration tasks. 51 00:03:40,339 --> 00:03:46,979 For more complex workflows, consider using LangGraph while still leveraging LCEL within individual nodes. 52 00:03:46,979 --> 00:03:51,380 As you develop your own applications, take advantage of LCEL's strengths, including 53 00:03:51,380 --> 00:03:57,339 parallel execution, async support, simplified streaming, and automatic tracing. 54 00:03:57,339 --> 00:04:02,059 These capabilities enhance both the power and maintainability of your applications. 55 00:04:02,059 --> 00:04:03,300 In this video, you learned that 56 00:04:03,300 --> 00:04:09,059 LCEL pattern structures workflows use the pipe operator for clear data flow. 57 00:04:09,059 --> 00:04:13,020 Prompts are defined using templates with variables and curly braces. 58 00:04:13,020 --> 00:04:16,540 Components can be linked using RunnableSequence for sequential execution. 59 00:04:16,540 --> 00:04:20,700 RunnableParallel allows multiple components to run concurrently with the same input. 60 00:04:20,700 --> 00:04:26,420 LCEL provides a more concise syntax by replacing RunnableSequence with the pipe operator. 61 00:04:26,420 --> 00:04:31,380 Type coercion in LCEL automatically converts functions and dictionaries into compatible components.