Request for information for a AI business app:

Hi everyone,

I'm working on a customer conversation transcription system for an EU-based organization, and I’d really appreciate some architecture guidance — especially around GDPR compliance, data handling in the EU, and integrating with Microsoft services like SharePoint, Azure, and Power BI.

Here’s what we need the system to do:

• Accurately transcribe customer conversations (ideally using two transcription engines for cross-checking)
• Store all output only in SharePoint, ensuring all data stays within the EU
• Log system behavior and performance without capturing any PII
• Provide Power BI dashboards for accuracy and compliance tracking
• Automatically evaluate transcripts for policy violations (e.g., use of prohibited language)

I’m mainly looking for help with the following design decisions:

1. Cloud vs On-Prem Transcription

What’s the best trade-off between compliance and scalability?

• Would a hybrid setup work, e.g., running Whisper locally for sensitive content and using Azure Speech Services (EU region) for the rest? I have read that these tools are not very accurate.

Or is it safer from a GDPR perspective to keep all processing fully on-prem?

2. Privacy-Friendly Logging

How can we configure Azure Monitor (or similar tools) to:

Track audio processing times and errors

Automatically redact personal info like phone numbers from logs

Integrate with SharePoint’s audit logging for traceability

3. SharePoint Integration

What are the best practices for:

• Automatically applying sensitivity labels to transcripts
• Enforcing retention policies (e.g., auto-delete after 90 days)

Extracting metadata like speaker names, timestamps, etc.

4. Analytics and Evaluation

How would you approach:

• Building a Power BI dashboard that shows transcription accuracy (e.g., Word Error Rate per agent)
• Flagging policy violations automatically
• Adding LLM-based evaluation (e.g., Azure OpenAI or another EU-hosted model) in a cost-effective way

5. End-to-End Compliance & Security

What’s the best way to enforce Zero Trust principles, including:

Conditional Access (e.g., block non-compliant devices)

• Data encryption using Azure Key Vault
• Using Microsoft Purview to support GDPR documentation and audits

6. Prototype Requirements:

• Data Flow: Audio → Transcription (dual engine for accuracy?) → SharePoint → Evaluation → Dashboard
• Needs to support:
  • Mobile use via an Intune-deployed PWA
  SSO via existing Azure AD
  • All data storage and processing within the EU

Docs I’ve Already Reviewed:

EU Data Boundary – Microsoft

Sensitivity Labels in SharePoint

Could anyone recommend a phased implementation plan or send me in the right direction? Ideally starting with a minimum viable product (MVP) that’s already compliant with GDPR?

Thanks so much in advance! I'd love to hear how others have tackled similar challenges, or if there's any Microsoft reference architecture that fits this use case.