How to build a cloud-based contact centre & analytics solution running mostly on AWS

Overview

This architecture describes a cloud-based contact centre & analytics solution running mostly on AWS, integrated with Azure Cognitive Services, and powered by several AWS Lambda functions to process customer/agent interactions, summarise data, and store insights.

It’s built to:

• Handle interactions between customers and call centre agents using Amazon Connect & Amazon Lex chatbot.

• Use GenAI Use Cases (UC1 & UC2) to generate summaries or analytics.

  • Stream and store data into various analytics & policy management platforms (EIS, EDAP).

Key Components and Flow

Customer & Agent Interaction

• Customers interact with the system via phone or chat.
• Agents (call centre operatives) handle calls via Amazon Connect.
• Amazon Lex Chatbot supports automated conversational interactions.
• A 3rd-party participant might also join the interaction.

Transcription & Data Streaming

After an interaction:

• Transcripts are published to a stream (aws-dub-genai-events) for processing.

• A DynamoDB Stream is used to validate the transcript.

Lambda Functions (UC1 & UC2)
Two GenAI Use Cases (UC1 & UC2):
- UC1: Customer-facing use case.
- UC2: Agent-facing use case.

They both are triggered by events from the stream and process transcripts or summaries through Lambda functions:
- python-genai-routing-processor-fcn
- python-genai-connect-cust-bot-summary-processor-fcn
- python-genai-connect-cust-agent-summary-processor-fcn

These Lambdas:

• Read prompts.
• Trigger summary generation.
• Store summaries (prompts & analytics) in S3 buckets and stream them to downstream systems.

Summary Feedback & Azure Integration

• Another Lambda (python-genai-cci-cust-bot-first-resp-fcn) stores the first response data and streams it back.

• Azure Cognitive Services is called via python-genai-azure-connector-fcn to perform further NLP or AI processing.

• The response from Azure is also stored & streamed back.

Data Storage & Streaming

• All outputs (summaries, analytics, feedback) are stored in S3 buckets.
• Data is also streamed to the EIS (policy management system), where agents can view/update customer policy details.

Periodic Analytics

• A time-based Event triggers another Lambda (python-genai-analytical-data-dump-fcn) to:
  • Read accumulated data.
  • Persist new records into an analytics datastore (edwh-genai-analytics).

Downstream Systems

EIS

• A Kubernetes-based Policy Management System, accessible via Amazon API Gateway.
• Stores policy & customer interaction details.

EDAP

• Ensures Data/Analytics Platform: another analytics and reporting platform.
• Data is loaded into EDAP from an S3 bucket (Load data into EDAP).

Azure Cognitive Services

• Connected via VPN & Azure Gateway for AI-enhanced cognitive functions.

Supporting Infrastructure

• Amazon S3 Buckets: Used throughout for data storage & streaming.
• Lambda Functions: Small serverless jobs that process, transform, and forward data.
• DynamoDB Stream: Ensures events are captured & processed.
• API Gateway & Kubernetes: For serving EIS as a web service.

Definitions in Diagram

Data Flow (Simplified)

• Customer contacts via Connect or Lex.
• Conversation transcript is captured & validated.
• Data streams to GenAI Lambdas for summarisation & analytics.
• Results stored in S3, streamed to EIS & EDAP.
• Azure Cognitive Services is optionally used for enhanced AI tasks.
• Periodic batch jobs persist analytics data for reporting.

Purpose of the Architecture

• Automates contact centre interactions with AI summaries.
• Provides agents with actionable insights on customer policies.
• Integrates multiple clouds (AWS & Azure) for cognitive AI tasks.
• Streams, stores, and analyses customer & agent data effectively.

Step-by-Step Scenario:

"A customer calls, and here’s what happens"

Step 1: Customer initiates contact

• A customer calls the contact centre.
• Amazon Connect handles the call.
• Depending on the situation, the customer might interact with:
  • Amazon Lex Chatbot (automated responses)
  • Or a live Agent (call centre operative).
• Sometimes a 3rd-party participant (like a supervisor) also joins.

Step 2: Transcript capture & validation

• As the conversation happens:
  • A transcript of the dialogue is captured and published to the event stream (aws-dub-genai-events).
  • DynamoDB Streams validate the transcript to ensure accuracy and readiness for processing.

Step 3: GenAI Use Cases kick in

• Two GenAI Lambda functions are triggered:
  • UC1: Creates a customer summary of the interaction.
  • UC2: Creates an agent summary (possibly focusing on agent performance or notes).
• Prompts and intermediate data are stored in S3 and streamed to other systems.

Step 4: Azure Cognitive Services

• If additional AI processing is needed (like sentiment analysis, intent detection, etc.), a Lambda (python-genai-azure-connector-fcn) sends the data to Azure Cognitive Services via a VPN.
• The response is read and stored back into S3.

Step 5: Data storage & feedback

• All outputs, summaries, prompts, and feedback are streamed back to:
  • EIS: So agents can see/update customer policy details.
  • S3 buckets: For audit and storage.
  • EDAP: For analytics and reporting.

Step 6: Periodic analytics dump

• On a schedule, another Lambda (python-genai-analytical-data-dump-fcn) runs.
• It reads accumulated analytical data, persists it into the analytics database (edwh-genai-analytics), and streams it downstream to EDAP.

Summary for a Presentation:

Modern Cloud Contact Centre with AI-driven Insights

Key Features:

• Customer-centric: Handles calls with Amazon Connect & Lex, enabling seamless customer-agent or bot interaction.
• Real-time transcription: Conversations are transcribed and validated instantly.
• AI-powered summaries: Two GenAI Use Cases generate customer & agent summaries using AWS Lambda.
• Hybrid Cloud AI: Leverages Azure Cognitive Services for advanced AI tasks over a secure VPN.
• Serverless & Scalable: Uses AWS Lambda for processing, S3 for storage, DynamoDB Streams for eventing.
• Analytics & Reporting: Streams data to EDAP and EIS for insights, dashboards, and policy management.
• Secure & Compliant: Policy management data is managed in EIS; sensitive analytics flow through controlled pipelines.

Benefits:

• Reduces manual note-taking for agents.
• Improves accuracy & speed of summarisation and reporting.
• Provides actionable insights for both customers & agents.
• Supports scalable analytics and compliance with minimal operational overhead.

Subscribe to my YouTube channel to see how I build this infrastructure