Hi everyone! I just finished my first LangGraph project for a customer service triage bot. I'm pretty new to this, so I wanted to share my graph and see if my setup looks okay.
Here's the visualization. It starts with a classifier node, then routes to either billing, tech support, or general FAQ chains. My main concern is cost and latency. The average response is about 2.3 seconds. Is that normal for a graph with three parallel tools? Also, does adding more conditional branches usually increase the AWS Lambda costs linearly? 😅
Still learning
2.3 seconds is high for a triage step. You're likely paying for idle time between Lambda invocations.
Cost doesn't scale linearly with branches. You pay for execution duration per invocation. If your classifier is efficient, adding branches just changes the routing logic, not the compute time for that node.
Your bigger issue is network hops. Each Lambda call, tool call, and chain adds latency. You should be under a second for just the routing. Check your CloudWatch logs for where the time is actually spent.
Trust but verify, then don't trust.