How microsoft/conversation-knowledge-mining-solution-accelerator works
Project type and purpose
• This is a full-stack “solution accelerator” that ingests, processes, and visualizes conversational data.
• Key goals:
• Transform audio/text conversation data into searchable indexes.
• Extract topics, key phrases, perform speech-to-text, and support interactive chat.
• Deploy end-to-end on Azure using App Service, Container Apps, AI services, databases, and search.
File structure to locate main components
• /src/App → React front-end (WebApp.Dockerfile, package.json, src/*)
• /src/api → Python API (FastAPI/Flask style app.py, api_routes.py, models, services)
• /infra → Bicep templates provisioning Azure resources (App Service, Cosmos DB, SQL DB, Storage, Key Vault, AI services, Container Registry)
• /infra/scripts → Data-processing scripts for search index, Azure AI Foundry (notebooks, Python, PowerShell, shell scripts)
• CI/CD: .github/workflows and .azdo/pipelines for build, test, deploy.
README for architecture insights
• The README contains a “Solution architecture” image showing high-level flows.
• Services used: Azure AI Foundry, Azure AI Content Understanding, Azure OpenAI Service, Azure AI Search, Storage, SQL, Cosmos DB, Key Vault, Managed Identity.
Main components
a. User Interface Layer
• React SPA deployed to Azure App Service (or Container App)
• Public web endpoint (HTTPS)
b. API Layer
• Python API container (App Service or Container App)
• Exposes REST endpoints: /api/chat, /api/history, /api/chart
c. Data Stores
• Azure SQL Database (structured metadata)
• Azure Cosmos DB (history, user sessions)
• Azure Blob Storage (raw audio/text files, intermediate data)
d. AI Services
• Azure AI Content Understanding (topic modeling, key-phrase extraction)
• Azure AI Foundry (processing pipelines, orchestration notebooks)
• Azure OpenAI Service (chat generation, embeddings)
• Azure AI Search (search index for processed conversation data)
e. Auxiliary Services
• Azure Key Vault (secrets, connection strings)
• Azure Container Registry (for Docker images)
• Managed Identity (secure access between services)
• GitHub Actions / Azure DevOps pipelines (CI/CD)
f. Data-Processing Scripts
• Azure CLI / PowerShell / Python scripts triggered post-deployment (index creation, data ingestion)
• Microsoft Fabric notebooks for advanced data prep (optional sub-component)
Relationships and interactions
1. User’s browser → HTTPS → React SPA
2. React SPA → REST calls → Python API
3. Python API:
• Reads/writes SQL & Cosmos DB via service clients
• Calls AI services:
• Content Understanding to extract topics/key phrases
• OpenAI for embeddings and chat responses
• Pushes processed data to Azure AI Search index
4. Data-ingestion pipeline scripts:
• Pull raw data from Blob Storage
• Use AI Foundry pipelines to process and output JSON
• Call Azure Search REST API to create or update the index
5. Deployment pipelines (GitHub Actions/Azure DevOps) → execute Bicep templates → provision infra → build & push Docker images → deploy App Service/Container App → run post-deployment scripts
Architectural patterns & design principles
• Separation of concerns: UI vs API vs Data Processing vs Infra provisioning.
• Microservices-style: API in its own container, UI in its own container.
• Infrastructure as Code: Bicep modules for repeatable, parameterized deployments.
• Managed Identity & Key Vault: no secrets in code, secure service-to-service auth.
• Eventual consistency: search index reflects processed data asynchronously.
Essential elements
• Clear labels: e.g., “Azure App Service (React)”, “Python API Container”, “Azure OpenAI: gpt-4o-mini”.
• A note highlighting optional advanced components (Fabric notebooks) but keeping primary focus on core solution accelerator.