QA, scorecards & human handoff

Keep your agents good and keep a human on hand for the hard cases. Mirakash scores every conversation with LLM-judge scorecards, flags issues by rule into a review queue — plus human-in-the-loop escalation that routes a hard case to a human provider or another agent and resumes the caller.

app.mirakash.com/qa
QA /Review queue
N
ScorecardsEscalationsIssues
Escalations
7
Failing
2
Avg score
4.6
Resolved
92%
#c8f2
Refund not resolved
+1 646 · billing
AMescalated
#c8e9
Low CSAT · 2.1
+1 312 · order status
PKopen
#c8d4
Agent stuck
+1 628 · reschedule
JDescalated
#c8c1
Wrong store hours
+44 20 · hours
AMopen
#c8b7
Off-policy answer
+1 210 · refund
PKreview
The problem

Agent quality drifts silently until customers complain

Nobody reviews ten thousand calls by hand, so bad answers slip through unseen, the same failure repeats, and there's no path for the one call the agent truly can't handle. The first sign of a problem is a churned customer.

How it works

The flow, end to end

01

Score

Each conversation is graded by an LLM judge against your scorecard the moment it ends.

02

Flag

Flagging rules open an issue on any call that breaches a threshold and queue it for review.

03

Escalate

A case the agent can't close routes to the human-in-the-loop queue for a provider or agent.

04

Resume

The human answers; the agent takes the answer back to the caller and continues the call.

Capabilities

What's in the box

LLM-judge scorecards

Every conversation is scored by an LLM judge against your rubric — no manual sampling.

Flagging rules

Rules flag conversations that breach a threshold and open an issue for review.

Review queue

Flagged calls land in one queue for a human to review, confirm, or dismiss.

Human-in-the-loop

Hard cases escalate to a queue that routes to a human provider or another agent.

Consult-and-resume

The human answers, and the agent resumes the caller — a consult, not a blind transfer.

Consult, don't transfer

Human-in-the-loop that answers the agent, not takes over the call

When an agent hits a case it can't close, it escalates to a queue that routes to a human provider or another agent, gets the answer, and resumes the caller itself — a consult-and-resume loop, not a blind transfer that drops the customer into a cold hold.

  • Escalations route to a human provider or another agent
  • Consult-and-resume: the caller stays with the agent
  • Every escalation is a tracked, resolvable record
app.mirakash.com/qa
QA /Review queue
N
ScorecardsEscalationsIssues
Escalations
7
Failing
2
Avg score
4.6
Resolved
92%
#c8f2
Refund not resolved
+1 646 · billing
AMescalated
#c8e9
Low CSAT · 2.1
+1 312 · order status
PKopen
#c8d4
Agent stuck
+1 628 · reschedule
JDescalated
#c8c1
Wrong store hours
+44 20 · hours
AMopen
#c8b7
Off-policy answer
+1 210 · refund
PKreview
Quality on autopilot

Score every call, flag the bad ones, review what matters

An LLM judge grades every conversation against your rubric, flagging rules open issues on the calls that breach a threshold, and a review queue puts exactly those in front of a human. What you learn feeds back into the agent — a flywheel, not a one-time audit.

  • LLM-judge scorecards on every conversation, no sampling
  • Rule-based flagging opens issues into a review queue
  • Findings feed back into the agent's instructions
app.mirakash.com/qa
QA /Review queue
N
ScorecardsEscalationsIssues
Escalations
7
Failing
2
Avg score
4.6
Resolved
92%
#c8f2
Refund not resolved
+1 646 · billing
AMescalated
#c8e9
Low CSAT · 2.1
+1 312 · order status
PKopen
#c8d4
Agent stuck
+1 628 · reschedule
JDescalated
#c8c1
Wrong store hours
+44 20 · hours
AMopen
#c8b7
Off-policy answer
+1 210 · refund
PKreview
100%
of conversations scored by an LLM judge
1
review queue for every flagged call
2
escalation targets: human or another agent
0
callers dropped by a blind transfer

Ready to put an AI agent on every call?

Book a walkthrough and see the voice pipeline, telephony, workflows, QA, and human handoff working together — on one platform.