Skip to main content

System Design GenBI

Problem Statement: GenBI — Unified Natural Language Business Intelligence Platform for Telecom​

The telecom enterprise maintains large volumes of operational, customer, and network data across multiple siloed systems—such as CRM (Customer Relationship Management), OSS/BSS (Operational/Business Support Systems), billing databases, network performance logs, and customer interaction platforms. Product and business analyst teams need quick insights spanning these heterogeneous data sources for decision-making, but current processes require technical mediation, SQL knowledge, and manual data joins, slowing down analysis and experimentation.

We aim to design GenBI, a Generative AI–powered Business Intelligence system, that allows analysts and product managers to query and visualize data in natural language while abstracting away data source complexity.

Core System Design Challenge​

How can we design a scalable, secure, multi-source conversational BI platform that:

  1. Connects to multiple databases (e.g., PostgreSQL, Snowflake, Redshift, MySQL, and NoSQL stores like MongoDB) across various telecom domains.
  2. Understands and translates natural language questions (e.g., “Show churn rate by region for the last quarter”) into optimized cross-source SQL or query plans.
  3. Resolves semantic differences (e.g., “subscriber” vs “customer_id”) via a unified semantic layer or knowledge graph.
  4. Ensures data governance and role-based access control while executing queries.
  5. Caches and optimizes frequent queries to maintain real-time response performance.
  6. Provides visual explanations and dashboarding capabilities based on LLM-generated outputs.

Key Design Goals​

  • Accuracy: Generate correct queries across multiple heterogeneous schemas.
  • Latency: Maintain sub-5-second query responses for common analytics.
  • Scalability: Handle data connections across multiple business units and terabyte-scale datasets.
  • Security: Enforce access policies and data masking in generated queries.
  • Explainability: Show how natural language was interpreted and translated into query logic.

Example Use Cases​

  • Product managers query: “What’s the ARPU trend for prepaid users in South India over the past 6 months?”
  • Analysts ask: “Compare customer complaints against dropped call rates for 4G networks.”
  • Executives request: “Summarize top churn reasons and affected revenue segments.”