[Suggestion] Amazon Connect real-time call transcription and analytics (Python)

[deepgram-robot]

What to build

A working example that integrates Deepgram real-time STT with Amazon Connect contact flows to provide live call transcription and post-call analytics for contact center agents. Should demonstrate streaming audio from Connect's media streams into Deepgram's WebSocket API with diarization for agent/caller separation.

Why this matters

Contact center developers building on AWS need a reference architecture showing how to wire Amazon Connect's audio streams to Deepgram for real-time transcription. This is one of the most common enterprise voice AI use cases — live agent assist, compliance monitoring, and post-call summarization all depend on accurate real-time transcription of phone calls. Developers currently have to piece together the integration from scratch with no reference implementation.

Suggested scope

• Language: Python (with CDK or CloudFormation for AWS resources)
• Deepgram APIs: STT streaming WebSocket (Nova-3), Audio Intelligence (summarization, sentiment)
• AWS services: Amazon Connect, Kinesis Video Streams (for media), Lambda
• Features: Real-time transcription with speaker diarization (agent vs caller channels), post-call summary generation, keyword/topic detection
• Complexity: Medium-high — requires AWS account with Connect instance

Acceptance criteria

• Runnable with minimal setup (clone, configure AWS + Deepgram API key, deploy)
• README explains the architecture and data flow clearly
• Uses current SDK version
• Demonstrates agent/caller speaker separation using diarization or multi-channel audio
• Includes post-call summary generation using Audio Intelligence features

---

Raised by the DX intelligence system.