InsightHub — AI-Powered Business Analytics Platform
InsightHub is a production cloud analytics platform integrating 10 Azure services end-to-end. A Python ETL pipeline downloads 7 CSV datasets from Azure Blob Storage, transforms and loads 230,000+ rows into a star schema (6 dimension tables, 3 fact tables) in Azure SQL Database. A Non-Clustered Columnstore Index on the 119,652-row FactSales table cut aggregate query time from ~8 seconds to under 1 second. A hybrid RAG search engine combines Azure AI Search (BM25 keyword retrieval + text-embedding-ada-002 vector search + semantic re-ranking) over 20 chunked internal business documents. GPT-4o runs in JSON mode (temperature 0.2, explicit schema) to generate structured business insights from live SQL metrics across four categories: Sales, Customers, Support, and Campaigns. The FastAPI backend implements JWT authentication, role-based access control (Admin / Analyst / Viewer), 14 REST endpoints, and Application Insights custom event tracking. The React 18 + TypeScript frontend has 6 dashboard pages including a guest demo mode. All 13 Azure resources are defined in Bicep IaC with Key Vault secret references — no plaintext credentials in config.
Tech Stack
Key Highlights
- Python ETL pipeline with data validation checks and documented source-to-target mappings; 230,000+ rows from 7 CSVs → star schema (6 dims, 3 facts) in Azure SQL
- Non-Clustered Columnstore Index cut FactSales aggregation from ~8 s to <1 s on 119,652 rows
- Hybrid RAG search: BM25 + ada-002 embeddings + Azure AI Search semantic re-ranking over 20 business documents
- GPT-4o structured JSON insight engine (temperature 0.2, JSON mode) reading live SQL metrics across 4 business categories
- FastAPI with JWT auth + RBAC (Admin / Analyst / Viewer), 14 endpoints, Application Insights telemetry
- Bicep IaC for all 13 Azure resources with Key Vault secret references — zero plaintext credentials in config