Skip to main content

5.3 Sequential Chains

AI-Generated Content

AI-generated content may contain errors. Always verify against official sources.

5.3 Sequential Chains

Key Concepts: Stage 1 output → Stage 2 input · Data transformation between stages

Official Docs: LangChain LCEL — Chaining

What is a Sequential Chain?​

A sequential chain passes the output of one step as the input to the next. Each step transforms the data.

Input → Step 1 (Extract) → Step 2 (Summarise) → Step 3 (Translate) → Output

Simple Two-Step Chain​

from langchain_openai import ChatOpenAI
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.output_parsers import StrOutputParser

llm = ChatOpenAI(model="gpt-4o-mini", temperature=0)

# Step 1: Summarise
summarise_prompt = ChatPromptTemplate.from_template(
    "Summarise the following article in 3 bullet points:\n\n{article}"
)

# Step 2: Translate the summary
translate_prompt = ChatPromptTemplate.from_template(
    "Translate the following text to {language}:\n\n{summary}"
)

# Chain them: article → summary → translation
chain = (
    {"summary": summarise_prompt | llm | StrOutputParser()}
    | translate_prompt
    | llm
    | StrOutputParser()
)

result = chain.invoke({
    "article": "Large language models are neural networks trained to predict the next token...",
    "language": "Arabic",
})
print(result)

Passing Multiple Values Between Steps​

Use RunnablePassthrough to carry values through a chain without modification:

from langchain_core.runnables import RunnablePassthrough, RunnableLambda
from langchain_core.output_parsers import JsonOutputParser

extract_prompt = ChatPromptTemplate.from_template(
    "Extract from this job posting as JSON with keys: title, skills (list), salary:\n\n{posting}"
)

validate_prompt = ChatPromptTemplate.from_template(
    """Given this job data: {job_data}
    Assess the role difficulty as: entry / mid / senior.
    Respond with JSON: {{\"level\": \"entry|mid|senior\", \"reason\": \"one sentence\"}}"""
)

# Step 1: extract structured data
extract_chain = extract_prompt | llm | JsonOutputParser()

# Step 2: classify from extracted data
classify_chain = (
    {"job_data": extract_chain}
    | validate_prompt
    | llm
    | JsonOutputParser()
)

result = classify_chain.invoke({"posting": "Senior ML Engineer at Acme. Python, PyTorch required. $180k/yr."})
print(result)   # {"level": "senior", "reason": "..."}

Adding Validation Between Steps​

Insert a RunnableLambda for validation:

from langchain_core.runnables import RunnableLambda
from pydantic import BaseModel, ValidationError

class JobData(BaseModel):
    title: str
    skills: list[str]
    salary: str | None

def validate_job(data: dict) -> dict:
    try:
        return JobData(**data).model_dump()
    except ValidationError as e:
        raise ValueError(f"Extraction failed schema: {e}")

chain = (
    extract_prompt
    | llm
    | JsonOutputParser()
    | RunnableLambda(validate_job)   # ← validation gate between steps
    | RunnableLambda(lambda x: {"job_data": str(x)})
    | validate_prompt
    | llm
    | JsonOutputParser()
)

Common Mistakes​

Common Mistakes
  1. Forgetting to pass all required keys — if Step 2 needs {language} but you only pass {summary}, you get a KeyError. Map all required variables explicitly.
  2. Chaining steps that depend on prior failures — always validate intermediate outputs. A failed extraction silently produces garbage that propagates through all downstream steps.
  3. Over-sequential chains — if steps are independent, run them in parallel (see 5.4) to reduce latency.

Quick Quiz​

Test Your Understanding

Q1. What LCEL component lets you pass an upstream value through to a later step without modification?
A1. RunnablePassthrough.

Q2. How do you inject validation logic between two LCEL chain steps?
A2. Wrap a validation function in RunnableLambda and insert it between steps with |.

Q3. What is the main risk of not validating between sequential steps?
A3. A failure in an early step propagates silently, causing all downstream steps to produce incorrect results without any obvious error.

Student Exercise​

Exercise 5.3 — News pipeline
Build a 3-step sequential chain: (1) summarise a news article in 3 bullets, (2) extract a list of named entities (people, organisations, locations), (3) classify the article category (politics, sport, technology, finance). Each step should validate its output schema before passing to the next.

Further Reading​

Next → 5.4 Parallel & Branching Chains