Idea
Declarative language automating predictive model training on relational databases to accelerate and scale machine learning workflows.
Research Paper
Core Innovation
This paper introduces Predictive Query Language (PQL), a SQL-inspired declarative language that unifies and automates the extraction of training labels for predictive modeling directly from relational databases. Unlike prior manual and error-prone methods, PQL enables defining diverse predictive tasks in a single query, supporting multiple machine learning paradigms and scalable implementations.
Why It Matters
Manual extraction of training data from relational databases is slow, error-prone, and labor-intensive, limiting the scalability of predictive modeling. PQL automates this process, reducing development time and mistakes, enabling faster deployment of predictive models across industries. This efficiency gain supports large-scale, diverse applications such as fraud detection and recommendation systems.
Market Size (TAM)
$20–50B TAM for predictive analytics and machine learning platforms; $2–10B SAM from financial services, e-commerce, healthcare, and supply chain sectors. Driven by growing AI adoption and demand for automated data workflows.
Potential Customers & Pain Points
- Financial institutions – Complex fraud detection data preparation
- E-commerce platforms – Laborious recommendation model training
- Healthcare providers – Difficult patient risk prediction data extraction
- Supply chain managers – Inefficient workload forecasting data setup
Business Model
Subscription-based SaaS platform offering tiered access to PQL tools and integrations, with enterprise licensing for large-scale deployments and custom support services.
Competitive Landscape
- DataRobot
- H2O.ai
- Google AutoML
- Microsoft Azure ML
- Databricks
Implementation Challenges
- Integration complexity with diverse database systems
- User adoption requiring familiarity with new query language
- Competition from established automated ML platforms
Validation Strategy
- Pilot deployments with financial and e-commerce clients to measure reduction in model development time
- Benchmarking predictive accuracy and label extraction efficiency against manual methods
- User feedback collection to refine language usability and integration features
Research Paper Overview
Predictive Query Language: A Domain-Specific Language for Predictive Modeling on Relational Databases
Summary
Predictive Query Language (PQL) is a SQL-inspired declarative language that simplifies defining predictive tasks on relational databases by automating training label computation for various machine learning tasks. It supports regression, classification, time-series forecasting, and recommender systems, and is integrated into a predictive AI platform with use cases in financial fraud detection, item recommendations, and workload prediction.