microsoft/Conversation-Knowledge-Mining-Solution-Accelerator

What it solves

Analysts often struggle to extract actionable insights from massive volumes of unstructured conversational data (transcripts, recordings, and documents). Manual review is time-consuming, inconsistent, and makes it is difficult to surface meaningful patterns or ask complex, contextual questions across the entire dataset.

How it works

The project implements an AI extraction pipeline that transforms raw data into an interactive analysis experience. It uses Azure Content Understanding and Azure OpenAI to extract entities, topics, and relationships. The processed data is indexed in Azure AI Search for hybrid retrieval and stored in Azure SQL Database for structured analytics.

A Microsoft Foundry ChatAgent acts as the primary interface, utilizing two tools: semantic search for retrieval and SQL for structured analysis. Additionally, an LLM analyzes the data schema to automatically plan and generate a dynamic KPI dashboard with charts and metrics based on the specific dataset uploaded.

Who it’s for

This solution is designed for organizational analysts—such as IT helpdesk analysts, loan analysts, and operations analysts—who need to mine sentiment, cluster recurring topics, and perform risk analysis across high-volume conversational logs.

Highlights

• LLM-planned dashboards: Automatically generates data-driven KPIs and charts by analyzing the dataset schema.
• Hybrid data exploration: Combines semantic search and SQL analytics through a single chat interface.
• ** multimodal data support**: Processes PDF, DOCX, JSON, CSV, TXT, images, WAV, and MP3 files.
• Flexible data ingestion: Supports direct file uploads or connecting to existing Azure AI Search indices and Microsoft Fabric Lakehouses/Warehouses.