Strongly Certified · Streaming Workflow

Call Center QA

Per-turn quality scoring. Full session archive. QA webhook on session end.

Voice support agent with programmatic quality scoring built into every turn. Audio and transcript are archived per session. The QA inbox gets a webhook when the call ends - score, flagged turns, full replay link.

Talk to an Engineer See What It Does

Per-turn

Programmatic eval-scorer

Full

Audio + transcript archived

On-end

QA-team webhook with score

What it does

The voice loop, end-to-end.

No black box. Each step is a typed-frame node you can edit, monitor, and replace.

The voice loop runs as a normal support session - STT, LLM, TTS over WebSocket.

After every turn, streaming-eval-scorer rates the response (tone, clarity, compliance). The score lands on the turn's span.

streaming-recorder writes the raw audio to S3. streaming-conversation-store writes the transcript and per-turn scores to your archive collection.

On session end, streaming-webhook-response posts the summary to your QA inbox - score, flagged turns, replay link.

Capabilities

Built for production. Day Two-ready.

Streaming graph contract, observability, and cost discipline come standard. The agent ships with a full test suite that runs in CI on every node version bump.

Per-turn evaluation

streaming-eval-scorer rates each agent reply against a rubric you define - tone, clarity, compliance, accuracy. Scores attach to the turn span and feed the session summary.

Rubric-drivenPer-turnOn span

Full session archive

Raw audio is archived to S3 by streaming-recorder. Transcripts and per-turn scores write to streaming-conversation-store. Replay any call, any turn, any time.

S3 audioTranscriptPer-turn rows

QA webhook on end

When the session ends, streaming-webhook-response fires a signed POST to your QA inbox with the session summary, score histogram, and flagged-turn IDs. Standard HMAC-SHA256.

SignedSession-endFlagged turns

Threshold flagging

streaming-conditional routes turns below your score threshold into a separate flagged collection. QA teams see only the calls that need review - not every transcript.

Threshold-drivenFlagged-only viewCost-friendly

Live span tree

Eval scores, recording offsets, and webhook delivery status all land as span attributes. The canvas overlay shows the call as it happened, with scores annotated per turn.

ADR-S14Score attributesReplay-ready

Compliance-tunable

Edit the eval-scorer rubric per-region, per-business-line, per-script. The same workflow runs against multiple scoring policies - pick one at deploy time.

Rubric per regionVersionedHot-swappable

Built on

Real services. Your stack.

Every dependency is a registered Strongly service or a model you control. Swap any one of them in the install wizard. The graph stays intact.

Eval scorer

streaming-eval-scorer - rubric-driven LLM judge

Recorder

streaming-recorder - S3 audio archive

Conversation log

streaming-conversation-store - transcript + scores

Webhook out

streaming-webhook-response - HMAC-SHA256

Five common customisations

Tune it. Don't fork it.

The marketplace template is the graph. Every customisation below is a config change or a single-node addition - never a rewrite.

Custom rubric

Edit the eval-scorer's rubric template. Common additions: brand-tone checks, regulated-disclosure verification, escalation-trigger detection.

Threshold-driven escalation

Add a streaming-handoff-detector after the scorer. Turns below threshold escalate to a human queue mid-call instead of post-call.

Selective archiving

Replace the always-on recorder with a flag-driven one - only flagged sessions get audio archived. Lower S3 spend, same QA coverage.

Compliance redaction

Insert streaming-pii-anonymiser before streaming-recorder so archived audio and transcript never carry raw PII.

Multi-rubric runs

Stand up the same workflow with different rubric configs (one per region or business line). Compare score distributions across deployments.