Expedia Group has introduced Service Telemetry Analyzer (STAR). It’s an AI-assisted internal observability platform that helps engineers investigate production incidents by analyzing service telemetry and generating structured root cause assessments. The system combines operational metrics and large-scale language models (LLMs) through predefined diagnostic workflows, aiming to reduce the time engineers spend identifying the causes of service degradation while putting human responsibility for validation and decision-making.
Rather than employing autonomous AI agents, the platform follows a deterministic workflow where telemetry is collected, analyzed using domain-specific prompts, integrated into intermediate results, and summarized into a final report with potential root causes and recommended next steps.
Regarding the project’s goals, the team writes:
The goal of this service was to minimize time to know (TTK) and time to recovery (TTR).
STAR Architecture (Source: Expedia Blog Post)
The platform is implemented as a FastAPI application that integrates with Datadog to obtain service metrics and an internally generated AI gateway that manages authentication and access to LLM providers. This workflow uses prompt chains to allow you to perform multiple specialized analyzes before performing integrated diagnostics. According to Expedia, the current implementation does not use features such as function calls, search augmentation generation (RAG), memory, or the use of autonomous tools, instead relying on predefined workflows to generate consistent analysis.
STAR focuses on standardized infrastructure telemetry collected from Kubernetes-based services and JVM applications. The system analyzes metrics such as request throughput, latency, HTTP, gRPC, and GraphQL error rates, CPU and memory usage, container restart events, Kubernetes readiness and liveness probe failures, Java heap usage, and garbage collection activity. Expedia explains that infrastructure metrics provide a consistent view across services developed using different programming languages and frameworks.
As the platform evolves, Expedia will continue to improve FastAPI background tasks. celery base An asynchronous architecture that uses Redis as both a message broker and results backend. The engineering team says that most processing consists of I/O-bound operations, including telemetry retrieval and LLM requests. The asynchronous execution model allows STAR to process multiple analytical tasks simultaneously while meeting the rate limits imposed by Datadog and our internal AI gateway.
The company reports that STAR is used to support production incident investigation, post-incident analysis, Kubernetes troubleshooting, and JVM memory diagnostics. Engineers review the results generated before acting on recommendations, making the platform an assistance tool rather than an autonomous operational system.
This blog also describes our ongoing work to extend the platform. Planned enhancements include the inclusion of service dependency information, additional operational metadata, Model Context Protocol (MCP)-based tool integration, and conversational interfaces. Expedia also evaluates STAR as part of its chaos engineering practice to help analyze the results of controlled failure experiments. Rapid management, tracking, and evaluation is now supported through Langfuse, and system performance is evaluated using subject matter expert reviews and user feedback.
