[Kyle Zhang (skywing918)]

fixes https://github.com/Azure/azure-sdk-pr/issues/2663
This pull request introduces a comprehensive Teams channel backfill capability to the Azure SDK QA Bot Agent, enabling the import of historical Teams messages into the bot's data store. The changes span documentation, configuration, backend models, and deployment pipelines to support this new functionality.

Key changes include:

Teams Channel Backfill Feature

• Added detailed documentation in README.md describing how to configure, deploy, and operate the Teams channel backfill, including authentication, job polling, and data verification steps.
• Introduced config/teams_collection_config.json to specify allowed Teams channels, their tenant keys, and paging limits for backfill operations.

Backend Model and API Enhancements

• Extended models/conversation.py with new models for Teams message attachments and images, and added request/response types for backfill job submission and status tracking. This supports richer message metadata and job lifecycle management.

Logic App and Deployment Pipeline Updates

• Updated the Logic App ARM template and deployment pipeline (pipelines/logicapp-cd.yml) to support a new teams-collection workflow, including environment-specific parameters and backend server identity configuration. [
• Added deployment parameter files for dev, test, and prod environments to control allowed channels and resource identities for the Teams backfill Logic App.

Data Model and Storage Changes

• Modified the Logic App template to store Teams message attachments in the extra_info field of conversation documents, ensuring parity with live message ingestion.

---

[azure-pipelines]

Azure Pipelines:
Successfully started running 1 pipeline(s).
70 pipeline(s) were filtered out due to trigger conditions.
There may be pipelines that require an authorized user to comment /azp run to run.

[Copilot]

🟡 Changes recommended

An update race can overwrite newer live edits with stale Teams data, while cached configuration can break initial backfills and coarse timestamp checks can retain stale summaries.

3 open issues

• Prevent overwriting newer edits with stale Teams payload · New
• Refresh backend configuration after publishing collection URL · New
• Avoid stale summaries when Cosmos timestamps are identical · New

What changed in this PR

Adds Teams channel history collection and scheduled Q&A summarization to the QA bot, supporting reusable evaluation content while keeping summaries separate from conversation messages.

Changes:

• Adds asynchronous backend backfill jobs with pagination, message normalization, and attachment/image preservation.
• Introduces a hosted summarization agent with validated Q&A output and freshness tracking.
• Extends deployment pipelines, infrastructure, configuration, documentation, and tests.

Paths below are relative to tools/sdk-ai-bots/azure-sdk-qa-bot-agent/.

File	Description
utils/​dashboard_identity.py	Uses Pydantic-compatible TypedDict.
utils/​azure_cosmosdb.py	Adds summary-container access and initialization cleanup.
tests/​test_teams_collection_service.py	Covers backfill, summarization, hosting, and deployment contracts.
services/​teams_thread_processor.py	Validates and versions model-generated Q&A.
services/​teams_qa_summary_service.py	Discovers threads and persists reusable summaries.
services/​teams_operations.py	Defines the summarization request protocol.
services/​conversation_service.py	Implements historical imports and background job tracking.
server.py	Exposes backfill submission and status endpoints.
scripts/​deploy_hosted_agent.py	Adds collection deployment and Routine management.
README.md	Documents deployment, identities, backfill, and verification.
pipelines/​teams-collection/​template.json	Provisions collection workflow, summary storage, and permissions.
pipelines/​teams-collection/​parameters.azure_sdk.test.json	Supplies test deployment settings.
pipelines/​teams-collection/​parameters.azure_sdk.prod.json	Supplies production deployment settings.
pipelines/​teams-collection/​parameters.azure_sdk.dev.json	Supplies development deployment settings.
pipelines/​logicapp/​template.json	Preserves attachments in live conversation requests.
pipelines/​logicapp-cd.yml	Adds collection deployment and backend identity resolution.
pipelines/​agent-cd.yml	Enables collection-agent deployment.
models/​teams_processing.py	Defines validated Q&A and resource-enrichment output.
models/​conversation.py	Adds attachment, image, and backfill job models.
config/​teams_collection_config.json	Configures channels, processing scopes, and schedule.
agents/​teams_collection_agent/​instructions.md	Defines thread selection and summarization rules.
agents/​teams_collection_agent/​init.py	Hosts background summarization and configures model tools.
agents/​teams_collection_agent/​Dockerfile	Packages the hosted collection agent.
agents/​teams_collection_agent/​agent.yaml	Registers the hosted agent name.

_{Review effort: Balanced}

[Jiaqi Zhang (JiaqiZhang-Dev)]

Besides the logicApp, I'm wondering if it's possible to use the Teams connection and call Graph API directly to get the channel's history messages?

[Kyle Zhang (skywing918)]

This is already how the backfill works: the backend requests message and reply pages from the existing Logic App, which uses the Teams connection to call Graph. The backend then normalizes the results and writes them to conversation-messages, using ETags to avoid overwriting newer edits. It does not call Graph directly.