Lcel And Runnables

Integrator composes prompt -> model -> parser by calling each step manually and passing intermediate values by hand instead of using the | operator.

Compose with the LCEL pipe: chain = prompt | model | StrOutputParser(). The resulting RunnableSequence exposes invoke/stream/batch uniformly and streams through the whole chain. Manual hand-wiring loses streaming, batching, and config propagation.

Integrator needs two independent sub-chains (a summary and a sentiment) over the same input but runs them sequentially, doubling latency.

Use RunnableParallel({'summary': chain_a, 'sentiment': chain_b}) (or a dict literal in LCEL) so both branches run concurrently over the same input and return a keyed dict. Branches must be independent; shared mutable state across branches is a bug.

A RAG chain must keep the original question while adding retrieved context, but the integrator overwrites the input dict and loses the question downstream.

Use RunnablePassthrough.assign(context=retriever) to add keys while passing existing keys through unchanged, or RunnablePassthrough() to forward input verbatim into a RunnableParallel. Preserve the original question for the prompt template.