You mentioned using Langgraph for error handling and recovery in LLM workflows. Can you explain how you apply prompt engineering techniques within these workflows to improve the model's performance in real‑world scenarios?

Langgraph is a framework for building LLM workflows with stateful nodes and directed graphs. In error handling, we wrap each node with try/except and use Langgraph’s built‑in retry and fallback mechanisms. Prompt engineering is applied by designing modular prompt templates that adapt based on the graph state. For example, we use a “fallback prompt” that asks the model to explain why it failed, then feed that explanation back into the next node to refine the request. We also use prompt templates that include context from previous steps, enabling the model to correct earlier mistakes. By combining Langgraph’s state persistence with dynamic prompt injection, we can recover from partial failures and improve overall accuracy in production.