Forum Discussion

roysampad91's avatar
roysampad91
Frequent Visitor
2 years ago
Solved

Is there any implementation (for ref.) of custom LLM models/Gen AI models hit by Power BI Reports?

Is there any implementation (for ref.) of custom LLM models/Gen AI models hit by Power BI Reports?

 

Understand using Smart Narrative and using the LLM Model API a connection can be estabilished theoritically but has anyone created a solution by connecting a custom LLM Model/Gen AI to Power BI?

 

Use Case: User uses Smart Narrative to ask a question which is passed to the Custom LLM model and the response is reverted back to the user on the Smart Narrative Visual.

 

Any reference implementations would be very helpful. Thanks in Advance.

  • Hi roysampad91 

    Integrating large language models (LLM) with Power BI is currently not very popular, due to several reasons:

    • Technical Complexity: Integrating an LLM with Power BI requires APIs, cloud services like Azure or AWS, and sometimes Power Automate or Azure Functions, making it complex and less accessible for many users.
    • Performance Costs: LLMs can add latency, which may slow down Power BI reports where users expect quick responses.
    • Limited Use Cases: Most Power BI users focus on traditional data analysis, so use cases for LLMs are narrow, mainly suited for scenarios needing detailed narrative explanations or language-based insights.
    • Existing Power BI Tools: Built-in features like Smart Narrative, Q&A, and AI Insights often fulfill users' needs, reducing demand for custom LLM models.
    • Privacy and Security Concerns: Using LLM models typically involves sending data to external servers, which may be a challenge for organizations with strict privacy requirements.
    • API Rate Limits: Some LLM providers have rate limits, which can restrict scalability for large user bases in Power BI.

    If you still want to explore LLM integration with Power BI, here are some options:

    If this post helps, then please consider Accepting it as the solution to help the other members find it more quickly.

2 Replies

  • Hi roysampad91 

    Integrating large language models (LLM) with Power BI is currently not very popular, due to several reasons:

    • Technical Complexity: Integrating an LLM with Power BI requires APIs, cloud services like Azure or AWS, and sometimes Power Automate or Azure Functions, making it complex and less accessible for many users.
    • Performance Costs: LLMs can add latency, which may slow down Power BI reports where users expect quick responses.
    • Limited Use Cases: Most Power BI users focus on traditional data analysis, so use cases for LLMs are narrow, mainly suited for scenarios needing detailed narrative explanations or language-based insights.
    • Existing Power BI Tools: Built-in features like Smart Narrative, Q&A, and AI Insights often fulfill users' needs, reducing demand for custom LLM models.
    • Privacy and Security Concerns: Using LLM models typically involves sending data to external servers, which may be a challenge for organizations with strict privacy requirements.
    • API Rate Limits: Some LLM providers have rate limits, which can restrict scalability for large user bases in Power BI.

    If you still want to explore LLM integration with Power BI, here are some options:

    If this post helps, then please consider Accepting it as the solution to help the other members find it more quickly.

  • Your approach is technically sound and aligns with how many are exploring this space. Using the Power BI REST API or XMLA endpoint to fetch metadata for context, then leveraging Azure OpenAI to interpret the question and generate the correct DAX or query is a solid path forward. The main challenges you'll likely face are structuring those prompts effectively and handling complex, multi-step questions reliably.

    In a similar scenario, I've also seen teams use a more direct integration where the natural language layer is embedded directly into the Power BI report itself. This can simplify the architecture by removing the need for a separate app to broker the conversation and query. The assistant interacts with the data model in real-time.

    If you're interested in that alternative approach, I shared a demo on LinkedIn showing an AI assistant built directly into a Power BI dashboard. It might give you another perspective on the user interaction piece: https://www.linkedin.com/feed/update/urn:li:activity:7438407183161065472