A natural language interface to Power BI, built on AWS

A FinTech business working with Cloud Combinator on AWS. The client is anonymised at their request.

The challenge had three focus areas.

An user's natural language question is sent to an Amazon Bedrock agent that is associated with a knowledge base populated with information about the Power BI semantic model. The agent converts the question into the equivalent DAX query. That query is handed to a workflow running on an Amazon EC2 instance, provisioned with the necessary Power BI connectivity, which executes the DAX against the model and returns the raw results. An AWS Lambda function then refines those results to match the phrasing of the original question, so the user receives a readable answer rather than a raw table.

The engagement delivered the GenAI question-and-answer workflow defined in the project success criteria to a proof-of-concept level: Lambda functions that accept an user query, use an Amazon Bedrock agent to convert it to DAX, execute it against the Power BI model and refine the answer, with Bedrock guardrails and agent memory in place. Acceptance was based on demonstrating the Q&An interaction and session continuity, and confirming the AWS environment is suitable for continued development.

• Natural-language querying — Employees can ask questions in plain English and receive answers refined from the Power BI model, no DAX knowledge required.
• Model-aware translation — An Amazon Bedrock agent backed by a knowledge base of the semantic model converts questions into DAX for execution.
• Self-correcting on errors — Failed DAX queries are fed back with their error text to be re-engineered and retried automatically.
• Safe and contextual — Amazon Bedrock Guardrails keep responses brand-compliant, and agent memory preserves conversation context across turns.
• Handover — A runbook, reference architecture and knowledge-transfer session were delivered so the client's team can operate and extend the system.

With the core natural-language-to-DAX capability proven, the client is positioned to harden the proof of concept towards production, adding user authentication and a front end, and to open self-service analytics to a much wider group of staff.