Idea
A graph-based multivariate time series dataset and platform for forecasting and anomaly detection in microservice systems.
Research Paper
Core Innovation
This paper introduces ChronoGraph, a unique dataset combining multivariate time series data with explicit service dependency graphs and expert-annotated anomaly labels from real microservice environments. Unlike prior datasets, it enables simultaneous evaluation of forecasting and anomaly detection tasks grounded in real incident data. This integration supports more realistic and effective model development for microservice performance monitoring.
Market Size (TAM)
$2–10B TAM, $1–2B SAM; assumption: growing cloud infrastructure monitoring and AI-driven incident management markets.
Potential Customers & Pain Points
- Cloud Service Providers Needing Reliable Microservice Monitoring
- DevOps Teams Struggling With Incident Detection
- AI Researchers Lacking Real-World Multivariate Time Series Benchmarks
Business Model
Subscription-based API access to the dataset and benchmarking platform with tiered pricing for enterprises and researchers.
Competitive Landscape
- Datadog
- Splunk
- New Relic
Implementation Challenges
- Data Privacy and Security Concerns
- Integration Complexity With Existing Systems
- High Expertise Required for Effective Use
Validation Strategy
- Pilot with cloud service providers for real-world forecasting accuracy
- Collaborate with DevOps teams to test anomaly detection effectiveness
- Publish benchmark results to attract AI research community adoption
Research Paper Overview
ChronoGraph: A Real-World Graph-Based Multivariate Time Series Dataset
Summary
ChronoGraph is a graph-structured multivariate time series dataset from real-world microservices, where nodes represent services emitting multivariate system metrics and edges encode service dependencies. It supports forecasting future performance metrics and includes expert-annotated anomaly labels for incident detection. This dataset uniquely combines multivariate time series, explicit dependency graphs, and real incident-aligned anomaly labels, enabling evaluation of forecasting and anomaly detection in microservice systems.