tencentmusic/supersonic
Supersonic
Supersonic est une plateforme d'analyse de données basée sur LLM et un moteur de couche sémantique qui traduit des questions en langage naturel en requêtes SQL exécutables. Elle fonctionne comme un tableau de bord de business intelligence et une interface texte-vers-SQL, permettant aux utilisateurs de récupérer des métriques métier et des insights via une interface conversationnelle.
Le système découple les définitions métier des schémas de base de données physiques en utilisant une couche logique gouvernée pour définir des métriques et des dimensions unifiées. Cette modélisation sémantique permet à la plateforme de mapper des modèles de langage humain vers des modèles sélectionnés et de traduire des déclarations sémantiques abstraites en SQL physique adapté à des moteurs de base de données spécifiques.
La plateforme fournit une passerelle de données d'entreprise dotée d'un contrôle d'accès granulaire et basé sur les rôles aux niveaux du jeu de données, de la colonne et de la ligne. Ses capacités incluent la gestion de dialogues multi-tours, la connectivité multi-bases de données et une architecture de plugins pour l'intégration d'outils tiers.
Le projet expose ses fonctionnalités via une API programmatique headless et une API de couche sémantique pour la consommation de données externe.
Features
- BI Semantic Layers - Provides a comprehensive BI semantic layer that maps raw warehouse data into governed business metrics and dimensions.
- LLM-Based SQL Generation - Translates natural language questions into executable SQL queries using large language models and semantic metadata.
- Natural Language Querying Interfaces - Provides a conversational interface that translates natural language inputs into structured data queries for business analysis.
- Text-to-SQL Translators - Implements a text-to-SQL translation layer that maps natural language questions to executable database commands.
- Natural Language Data Analysis - Translates human language questions into data queries to retrieve business metrics and insights from databases.
- Conversational Data Analysis Tools - Provides a conversational interface that translates natural language questions into SQL to analyze business metrics.
- Natural Language Querying - Translates human-language questions into executable data queries and corresponding visual charts.
- Natural Language to SQL - Transforms complex natural language queries into structured SQL using language parsers and semantic interpreters.
- Semantic Data Models - Defines a unified layer of metrics and dimensions to create consistent business definitions across diverse data sources.
- Semantic to SQL Translation - Translates abstract semantic query statements into physical SQL tailored to the target database engine.
- SQL Translation Layers - Converts abstract semantic query statements into executable SQL tailored for specific physical data models.
- Governed Logical Layers - Defines a governed logical layer of metrics and dimensions to decouple business definitions from physical schemas.
- Role-Based Access Control - Restricts access to specific data domains and AI agents based on assigned user permissions.
- Row and Column Access Controls - Enforces granular security by restricting data visibility at the dataset, column, and row levels.
- Dialogue Context Management - Maintains conversation history and context across multiple turns to refine data queries and improve analysis accuracy.
- Example-Based Prompt Memory - Improves query accuracy by storing and retrieving example prompt-response pairs for the language model.
- Business Intelligence Dashboards - Converts natural language inputs into interactive charts and data-driven answers via a BI dashboard.
- Data Attribute Search - Locates specific dimensions and metrics using natural language, fuzzy matching, or predefined aliases.
- Data Gateways - Provides a secure access layer and unified interface for retrieving consolidated data from multiple disparate sources.
- Semantic Data APIs - Exposes unified data models and business definitions through a specialized API for consumption by external tools.
- Multi-Database Connections - Integrates with multiple database engines to consolidate data from various storage systems into a single interface.
- Semantic Schema Mapping - Maps user text to specific schema attributes using fuzzy matching and predefined aliases within a semantic index.
- Derived Metrics - Allows the creation of new metrics based on existing measures or field expressions to expand analysis.
- Business Metric Mappings - Maps natural language patterns to curated semantic models to generate data visualizations.
- Agent Configurations - Manages model connections and domain knowledge using term identification and few-shot memory configurations.
- Metric Element Mapping - Matches user query text against a knowledge base of schema elements to identify referenced business metrics.
- SQL Enhancement - High-performance SQL generation tool for complex analytical queries.
