Shraddha Yeolekar
← All projects
Project 03 · Agentic AI

Multi-Agent Support Orchestrator

AgenticHuman-in-the-loop

A support bot where a router agent reads each message and decides whether a knowledge-base agent answers it or a human takes it.

Key decision
Upset customers and exception requests go to a person, not the bot.
Why
That's where a fluent wrong answer does the most damage: granting a refund the policy doesn't allow, or arguing with someone who is already angry. The escalation agent doesn't reply to the customer. It writes a two-sentence summary with an urgency label, so the person picking it up starts with context.
Gave up
Some tickets the bot might have handled now go to the human team.
Customermessage RouterKB or ESCALATE policy question Vector searchbest-matching fact KB agentanswers from it → reply to customer angry or exception Escalation agentsummary + urgency Human agentpicks up ticket every model call is traced: latency, tokens in and out
The highlighted path is the decision: the riskiest conversations skip the bot's answer and reach a person with a summary.