Walkthrough Complete - AI Contact Centre
You’ve completed the AI Contact Centre (Council Reception) walkthrough
Great! You've deployed the demo
Now let's walk through what you just deployed and see it in action.
Start WalkthroughChoose your next step
Generate Evidence Pack
Create your business case documentation with what you've learned.
Generate Evidence PackWalkthrough complete
In about 38 minutes you dialled a real UK number from your own phone, sent a photo through the in-browser simulator, called the same bot from your laptop with no phone at all, took a safeguarding call as a Connect agent, edited the bot's greeting in the GUI, watched the bot triage a distressed multi-intent caller, and heard it answer in Italian.
What you've learned
📞 Real phone line, real bot
Amazon Connect claimed a real UK phone number. You dialled from your own mobile or landline, and an AWS-hosted bot answered. Lex v2 classified the intent and a Bedrock Knowledge Base answered with citations.
Value: end-to-end voice on the AWS stack, no third-party telephony
🧠 Multi-intent triage
A layered distressing call produced four distinct council-service intents, each acknowledged by name. Bedrock Guardrails blocked hallucinated legal advice. Sentiment chips climbed from green to red as distress rose.
Value: handles real callers' real lives, not one issue at a time
📷 Multimodal photo description
A photo uploaded through the in-browser simulator was classified by Bedrock Nova Pro into a structured object class, severity, and suggested council action, all attached to a Connect Case automatically.
Value: residents can show, not just tell, and the system understands
🔗 Cross-channel single case
When the caller dialled back, Customer Profiles recognised the phone number, the contact flow looked up the open case, and the bot read back a single sentence acknowledging both the call and the photo, with a shared reference number.
Value: one resident, one case, no duplicates across channels
🛡 PII redaction at source
Contact Lens redacted names, addresses, and phone numbers at the Connect instance level. Only redacted segments are saved in DynamoDB and the share PDF. Bedrock Guardrails also anonymise model output.
Value: sensitive content never leaves the Connect boundary in the clear
🖥 Companion web app
A three-pane web app showed the live transcript, the case being built, and the photo simulator. The same web app also offers a real Connect chat session and browser-based voice, video, and screen sharing through the Chime SDK.
Value: see what's happening end-to-end on one screen
🎧 Agent hand-take
Marking yourself Available in the agent workspace and accepting an inbound call gave you a live transcript, sentiment chips, and a safeguarding flag flipped on the case as soon as sustained distress was detected. The redacted transcript is what the agent sees, even live.
Value: humans in the loop, with full context, no PII exposure
✏️ Editable in the console
You changed the bot's greeting in under a minute by clicking into the contact flow in the Connect admin GUI. No deploy, no developer. The change is live until the next IaC deploy resets it.
Value: customer services teams can iterate copy without a release cycle
🌍 Six locales, one bot
Lex transcribes en, it, fr, de, es, pl with neural Polly voices. Welsh and Romanian fall back to en_US ASR with Polly playback in the target language. Locale flips per contact based on Bedrock's language detection from the caller's first reply.
Value: residents heard in their own language with no menu and no transfer
Where to go next
- Try the Council Chatbot scenario for a Bedrock-grounded chat-only experience.
- Try the Text-to-Speech scenario for accessibility audio generation.
- Try the FOI Redaction scenario for a deeper look at PII handling.