Build SimpleSequentialChain Yourself

Some tasks genuinely benefit from being broken into stages — extract structured data first, then reason over it, rather than asking a single prompt to do both at once. Sequential chaining is the mechanism that makes that decomposition practical: each stage stays simple and independently testable, while the pipeline as a whole handles the full task.

There's no hidden magic in sequential chaining — it's a plain for loop threading one variable through each step. That simplicity is a feature: because it's just Python control flow, it's trivial to reason about, debug (print the intermediate value between steps), and extend with additional chains.

A single chain handles one step. Real reasoning is often multi-step: extract a keyword, THEN write about it. LangChain's SimpleSequentialChain wires multiple chains together so the output of one becomes the input to the next, automatically.

The mechanism is a simple loop: start with the initial input, then for each chain in sequence, run it and let its output become the next chain's input. Whatever comes out of the last chain is the final result.

Build SimpleSequentialChain Yourself. extract_chain pulls out a keyword; tweet_chain writes about whatever keyword it's given. Finish SimpleSequentialChain.invoke(): run each chain on the current value, in order, and feed each result forward as the next chain's input.

In a SimpleSequentialChain with 3 chains, what is passed as the input to the third chain?

• →The output of the second chain — each chain's output becomes the next chain's input, not the original initial_input.
• →The original initial_input, unchanged, passed to every chain in the sequence.

You've now built two different chain composition styles. Modern LangChain expresses this same idea more elegantly with LCEL — the pipe operator. Next lesson builds that operator overload yourself.