LangChain & LCEL (Expression Language)

1. Overview#

LangChain is an orchestration framework for developing applications powered by Large Language Models. LangChain Expression Language (LCEL) is a declarative method to compose arbitrary chains from primitives (PromptTemplate | ChatModel | OutputParser), providing first-class support for streaming, async execution, batching, and fallbacks.

---

2. The LCEL Composition Paradigm#

›PromptTemplate ──► ChatModel ──► StrOutputParser
 (Inputs) (Inference) (Clean Text)

With LCEL, the pipe operator | chains components together seamlessly:

🐍 PythonInteractive WebAssembly
EditCopyAsk AIRun Code
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.output_parsers import StrOutputParser
from langchain_openai import ChatOpenAI

# 1. Define Prompt Template
prompt = ChatPromptTemplate.from_messages([
 ("system", "You are an elite data scientist. Explain the concept in 2 crisp bullet points with code."),
 ("human", "Explain {concept} in Python.")
])

# 2. Define Model
model = ChatOpenAI(model="gpt-4o-mini", temperature=0.2)

# 3. Define Parser
parser = StrOutputParser()

# 4. Compose LCEL Chain
chain = prompt | model | parser

# 5. Execute Chain
result = chain.invoke({"concept": "Vector Embeddings"})
print(result)

---

3. Streaming and Batch Capabilities#

LCEL chains expose built-in asynchronous and streaming interfaces without changing application code:

🐍 PythonInteractive WebAssembly
EditCopyAsk AIRun Code
# Streaming token-by-token
for chunk in chain.stream({"concept": "Cross-Validation"}):
 print(chunk, end="", flush=True)

# Parallel batch processing
batch_inputs = [
 {"concept": "Precision vs Recall"},
 {"concept": "ROC-AUC"},
 {"concept": "F1-Score"}
]
batch_results = chain.batch(batch_inputs)

---

4. Best Practices Checklist#

• Use RunnableParallel or RunnablePassthrough when routing multi-source context (e.g. Question + Retrieved Documents).
• Implement .with_fallbacks([backup_model]) to guarantee high availability during upstream LLM outages.
• Use LangSmith tracing for observability and token cost monitoring in production.