Source: Build LLM Apps tutorial (Notebook)

The owner's tutorial on LangChain, LangGraph and LangSmith, including zero-code LangSmith tracing and a sample run tree with per-step latency, tokens and cost.

Contents
  • What: interactive HTML tutorial in the owner’s Notebook
  • Author: Loic (coolloic)
  • Date: accessed 2026-11-14
  • Original: Build LLM Apps

Summary

Frames app building around three problems: glue code (LangChain), non-linear logic (LangGraph) and the black box (LangSmith). The LangSmith section shows tracing and a small evaluation script.

Key claims

  • Setting LANGSMITH_TRACING=true and an API key traces LangChain and LangGraph apps with no code change.
  • A trace is a nested run tree; the example agent run shows each LLM call and tool call with its own latency, token count and cost, making the reason-act-observe loop visible.
  • Tracing answers “which step failed”: bad retrieval, weak prompt, or the model.
  • LangSmith evaluation runs a target function over a dataset with grader functions to catch regressions before shipping.

Feeds

LLM observability

Suggest changes →