Interview: LangChain and LangGraph

What you will be able to do

• ✓Explain what LangChain provides and what LCEL is
• ✓Explain LangGraph state, nodes, edges and reducers
• ✓Describe cycles, conditional edges and recursion limits
• ✓Explain checkpointing, threads and human-in-the-loop
• ✓Choose between a chain, a graph and no framework

The idea, in plain English

Many job descriptions name LangChain and LangGraph, so expect specific questions: how state works, what a reducer is, how a graph loops and stops, and how to pause for a human.

Interviewers also test judgement: why use a framework at all, and when is LangGraph better than a plain chain?

Modules 2 and 3 built every piece; use them as examples.

Worked example: Fourteen framework questions answered with the graphs built in Module 3 and the chains of Module 2.

Basic: What is LangChain?

A Python (and JavaScript) framework of building blocks for LLM apps: model wrappers, prompt templates, output parsers, tools, retrievers and a way to chain them (LCEL).

• •Standard interfaces let you swap models and stores.
• •Lesson 2.1 discusses why a framework helps - and what it costs.

Basic: What is LangGraph, and why does it exist?

LangGraph builds agents as graphs: shared state, nodes that update it, and edges that decide what runs next - including loops. It exists because agents need cycles, branching, persistence and human pauses, which simple chains do not handle well.

• •Chains are straight lines; agents loop.
• •Lesson 3.1: why graphs, not chains.

Basic: What are state, nodes and edges in LangGraph?

State is a typed dictionary shared by the graph. A node is a function that reads the state and returns updates. An edge says which node runs next - fixed, or chosen by a function (a conditional edge).

• •Nodes return partial updates, not the whole state.
• •Lesson 3.2.

Intermediate: What is a reducer in LangGraph state?

A reducer decides how a node's update is combined with the existing value - for example add_messages appends new messages instead of replacing the list.

• •Without a reducer, an update overwrites the field.
• •Important for message history and parallel branches.

Intermediate: How do conditional edges work?

A conditional edge calls a routing function on the current state; its return value picks the next node - for example "tools" if the last message has tool calls, otherwise END.

• •Keep routing functions simple and deterministic.
• •Lesson 3.3.

Intermediate: How do you stop a LangGraph agent from looping forever?

Count steps in the state and route to an end or give-up node when a limit is reached. LangGraph's recursion_limit is a final safety net that raises an error when too many steps run.

• •Your own limit gives a clean answer; the recursion limit gives an exception.
• •Lesson 3.4.

Intermediate: What are checkpointers and threads?

A checkpointer saves the graph state after every step. A thread_id names one conversation, so the graph can continue it later, recover after a crash, or be inspected and replayed.

• •In-memory checkpointers for tests; database-backed ones for production.
• •Lesson 3.6.

Intermediate: How does human-in-the-loop work in LangGraph?

The graph interrupts before (or inside) a node, saves its state with the checkpointer, and waits. A person reviews or edits, then the graph resumes from the saved state.

• •Needs a checkpointer, because the process may wait minutes or days.
• •Use it before risky actions. Lesson 3.7.

Intermediate: What streaming options does LangGraph have?

You can stream state values after each step, only the updates, or tokens from the model as they are generated - so users see progress instead of waiting.

• •Token streaming improves perceived speed.
• •Step streaming helps debugging. Lesson 3.9.

Advanced: How do you build a multi-agent system in LangGraph?

Make each agent a node or subgraph and connect them with a supervisor node that routes work, or with hand-offs where one agent passes control to another. Shared state carries messages and results.

• •Keep each agent's tools and prompt focused.
• •Watch total calls and cost. Lesson 3.8.

Advanced: When would you not use LangChain or LangGraph?

For a simple app, plain code with the model's SDK can be clearer, faster and easier to debug. Frameworks add abstraction, version churn and learning cost; they pay off for complex agents, persistence and integrations.

• •Module 1 built agents with no framework at all.
• •Know what the framework does underneath.

Advanced: What is LCEL, and what are its benefits?

The LangChain Expression Language composes components with the | operator into runnables that share one interface - invoke, batch and stream - with built-in parallelism and tracing.

• •prompt | model | parser.
• •Every runnable can be streamed or batched the same way. Lesson 2.4.

Advanced: How would you test a LangGraph agent?

Unit-test nodes as plain functions with sample state, test routing functions with crafted states, replace the model with a fake in tests, and run end-to-end evaluations on real tasks with tracing.

• •Deterministic parts get normal tests; model behaviour gets evaluations.

Advanced: How do you deploy a LangGraph agent?

Build the graph once at startup, serve it behind an API (for example FastAPI), use a persistent checkpointer, set timeouts and concurrency limits, and trace runs in production.

• •Lesson 5.7 packages an agent as a command and a web service.

Interview questions