What this is

No new concepts in this lesson. This is a checkpoint: a small, real pipeline built entirely out of ideas from Lessons 1 through 6, combined into one thing. If you can read the code and explain why every @traceable, run_type, tag, and thread_id is there, you've mastered the Beginner tier. If any piece feels unfamiliar, revisit the lesson it came from before continuing to Intermediate.

What it does

Runs three product reviews through a pipeline (clean text, classify sentiment with Gemini, count words), fully traced, with every review in this run grouped into a single thread in the LangSmith UI.

Where each piece came from

@traceable(run_type="tool", tags=["beginner-checkpoint"])
def clean(text: str) -> str: ...


@traceable(run_type="tool", tags=["beginner-checkpoint"])
def word_count(text: str) -> int: ...

Lesson 3: run_type="tool" for small, deterministic helper functions, plus a tag identifying this project's runs.

@traceable(run_type="chain", tags=["beginner-checkpoint"])
def classify_sentiment(text: str, batch_id: str) -> str:
    response = model.invoke(text, config={"metadata": {"thread_id": batch_id}})

Lesson 3 again (run_type="chain" around an auto-traced Gemini call, producing a nested "llm" run) and Lesson 5 (thread_id passed through config, so this nested call's run also carries the batch's thread id).

@traceable(run_type="chain", tags=["beginner-checkpoint"], metadata={"lesson": 7})
def process_review(review: str, batch_id: str) -> dict:
    cleaned = clean(review)
    return {"word_count": word_count(cleaned), "sentiment": classify_sentiment(cleaned, batch_id)}

Lesson 2: three separately traced functions called from inside process_review, becoming its children in the run tree.

batch_id = str(uuid.uuid4())
...
result = process_review(review, batch_id, langsmith_extra={"metadata": {"thread_id": batch_id}})

Lesson 5: one batch_id shared across all three calls to process_review in this run, passed via langsmith_extra so all three top-level runs (and everything nested under them) join the same thread.

Checkpoint

  • Every review is cleaned, classified, and counted, fully traced start to finish.
  • All three reviews in a run share one batch_id, so the Threads view groups them as one session.
  • Run types, tags, and metadata layered together give each run both a category and searchable labels.

Try this yourself, without looking anything up: add a fourth review to the list, does it join the same thread with no other changes? Change one tag on just classify_sentiment, can you find it later by filtering on that tag? Remove the thread_id from one call, and confirm that run now shows up outside the thread, on its own.

If you can make these changes confidently, you're ready for the Intermediate tier, starting at Lesson 8.