Sugar BI: AI‑Driven Next‑Generation Business Intelligence Platform

This article summarizes the presentation "New Generation BI Platform Driven by Large Models – How to Enable Smart Decision‑Making" from the 2024 Baidu Create Conference.

It outlines the evolution of Sugar BI, originally the internal reporting platform ShowX, which was rebuilt and rebranded in 2018. Since then, Sugar BI has released versions 2.0 and 3.0, expanding visual chart types, data‑source support, and third‑party integrations, and has grown to over 10,000 daily active users.

From 2021 onward, Sugar BI began exploring intelligent analysis, embedding AI capabilities into the product. Version 4.0 introduced advanced analytics such as automatic analysis and attribution analysis. In the era of large‑model and generative AI, Sugar BI launched the "Sugar Bot" intelligent Q&A module, reducing multi‑day data‑analysis tasks to a few minutes.

Key technical highlights include:

Support for more than 30 data sources, covering relational databases (MySQL, PostgreSQL, Oracle, SQL Server, etc.), Chinese‑domestic databases (达梦, 人大金仓, 华为 GaussDB), big‑data engines (Hive, Spark, Doris, ClickHouse, Presto), NoSQL and time‑series stores (InfluxDB, Prometheus, Redis, MongoDB), as well as Excel/CSV uploads and API connections.

A zero‑code, drag‑and‑drop visual editor built on Baidu‑open‑source ECharts and D3, offering over 150 chart components, 3‑D visualizations, and ready‑made dashboard templates for various industries.

Data‑model layer that automatically distinguishes dimensions and measures, supports multi‑table joins, virtual calculated fields, and adapts SQL dialects across heterogeneous databases.

BI engine providing filtering, drill‑down, cross‑chart interaction, and second‑stage calculations (e.g., retention analysis, pivot tables, alert calculations).

Intelligent chart recommendation that extracts chart features from user‑dropped fields, scores candidate charts using mandatory (hard) and optional (soft) rules, and automatically selects the most suitable visualization.

Sugar Bot workflow: prompt engineering → NL2JSON conversion → intervention & correction → intelligent chart generation. The system leverages Baidu’s 文心 large models via the Qianfan platform, few‑shot examples, private‑domain knowledge, and function‑call support for chart‑type specification.

The presentation also details the architecture of Sugar Bot, including prompt components (field schema, example data, output format, notes, few‑shot examples, user query) and the feedback loop that uses user likes/dislikes to generate better few‑shot prompts and fine‑tune the model.

Overall, Sugar BI integrates AI, BI, and visualization into a containerized SaaS platform, enabling rapid, code‑free data analysis and smart decision‑making for enterprises.