Hi everyone,
I’m working on designing a bot experience in Microsoft Teams that allows users to query report data using natural language. Below is the intended behavior — I’d appreciate feedback or suggestions on improving this approach.
Core Behavior
1. Teams Integration
The bot is available within Microsoft Teams.
2. Personalized Greeting
When a user opens the bot, it greets them based on:
Time of day (e.g., Good Morning / Afternoon / Evening)
Their Microsoft Entra ID display name
3. Natural Language Query
Users can ask questions in plain English, for example:
“Get the count of studies which are Phase 1”
4. Intelligent Report Identification
The bot should:
Respect user access
Only consider reports the user has access to
Example: ABC, CHF, BCA, ACF
Understand user intent (semantic matching)
Identify key entities/attributes from the query
Example:
Query: “Get the count of studies which are Phase 1”
Key concepts: Studies, Phase 1
Map query to relevant reports
Determine which reports contain the required data
Example:
Relevant reports: ABC, CHF
Prompt user for report selection
Respond with something like:
“I found relevant data in the following reports. Please select one to proceed.”
Require the user to select exactly one report
Note:
The bot should infer intent even when exact attribute or column names are not used — similar to how Power BI Copilot interprets queries.
5. Query Execution
Once the user selects a report:
The bot answers the question using only that report’s semantic model
Ensures accuracy and avoids cross-report ambiguity
Goal
To create a Copilot-like experience within Teams that:
Understands natural language queries
Respects user-specific data access
Guides users to the correct data source
Provides accurate, context-aware responses
