The author describes how trivago repurposed a subset of its existing end-to-end Cucumber/Selenium scenarios, using Gherkin tags, for three new jobs: synthetic production monitoring, per-release browser performance measurement, and stage stability checks after backend changes. The post highlights practical lessons such as avoiding greenwashing with meaningful assertions, keeping tests side-effect free, and recycling selectively to avoid overloading the stage environment.
trivago engineering blog
Loading
trivago's engineering team describes how it runs 600+ short-lived preview environments on GKE, one per pull request, using KEDA's HTTP add-on for scale-to-zero with wake-on-request. The post explains why Knative and cron-based scaling were rejected, and how a second commit-specific URL was added after a production incident where rolling updates masked a crashing pod. It also covers the trade-offs of putting a proxy on the request path.
The post explains why LLMs are stateless and how conversational memory is built in the application layer. It compares resending full history, RAG-based memory, and agent frameworks, then walks through the SimpleMem three-stage pipeline of compression, consolidation, and adaptive retrieval. It also covers a trivago chatbot demo with token-cost measurements and a proposed use of similar techniques for query-intent matching in hotel search.
The trivago UX team details how AI agents and orchestrated workflows transformed copywriting, localization, research, and product design. The team reports 75% faster copy turnaround, reduced prototyping time, reusable domain systems, and lessons for designing AI-native workflows with human quality controls.
This report summarizes three QA talks covering AI-adoption measurement with commit attribution and DORA metrics, randomized test inputs with reproducible seeds, and an audited AI agent for API integration testing. It highlights practical safeguards including structured traces, deterministic reproduction, uncertainty reporting, and human review.
trivago recaps its 2026 Tech Get Together, highlighting the company’s growing adoption of AI, agentic workflows, and frictionless engineering practices. The event brought together more than 200 technology professionals for talks, knowledge sharing, and discussions about evolving technical roles and organizational collaboration.
trivago details a layered investigation into Kafka consumer lag, slow polling, reactive backpressure, gRPC thread-pool limits, and in-memory delays. By replacing the custom consumer, removing the delay, and tuning concurrency, the team cut pod count and deployment costs by 83%, eliminated lag alerts, and reduced startup time from about 60 seconds to 10.
Trivago explains how it uses an AI-powered GitHub Actions job to analyze persistent end-to-end test failures. The workflow combines rerun results, cross-platform correlation, code diffs, and recent test history to identify flaky tests, infrastructure issues, and likely regressions while controlling latency and LLM costs.
The article examines trivago’s six-year effort to unify fragmented internal APIs with a GraphQL Mesh gateway for administration tools. It details schema composition, transforms, access control, audit logging, environment complexity, and the case for schema registries, while considering the gateway as a foundation for AI-assisted internal tooling.
Trivago explains how it extended an axe-core and Selenium-based accessibility framework with baseline comparison, stable violation fingerprints, and GitHub Actions integration. The approach separates new, fixed, and persisting issues so teams can catch regressions in pull requests while tracking accessibility improvements over time.
Armin Aminian explains how trivago replaced always-on Kafka sink deployments with KEDA-based, scale-to-zero workloads driven by consumer lag. The post covers configuration, partition and threshold tuning, consumer-group cleanup, staged rollout practices, and the resulting reduction from roughly 50 replica-hours to 1–2 per region daily.
Trivago describes how an Appium-based mobile test framework evolved from a learning project into a maintainable solution for iOS and Android. The rewrite adopted Appium 2/3’s driver model, parallel threads, lifecycle-aware plugins, shared testability practices, and end-to-end accessibility checks.
Anis Khan explains how service-level objectives and error budgets can replace opinion-driven tension between developers and product managers with shared, measurable reliability goals. The post covers SLO calculations, alerting, burn rates, and trivago’s use of availability, latency, and business metrics to guide feature delivery.
Dmytro Kurets explains how trivago’s mixed backend/frontend team used structured mob programming to reduce review bottlenecks and improve collaboration. The post covers driver rotation, remote tooling, onboarding, architecture work, migrations, and the conditions under which mobbing outperforms solo development.
Trivago details how it scaled AI from isolated experiments to company-wide adoption between 2023 and 2025. The post shares adoption and productivity metrics, an AI ambassador program, governance processes, and examples including an IT support chatbot and internal AI agents.
The authors describe how trivago’s QA team developed and consolidated custom Chrome extensions to streamline exploratory testing, debugging, and data investigation. The post covers the move to Manifest V3 with TypeScript and React, plus a GitHub Actions-based distribution workflow and the limitations of auto-updating privately installed extensions.
Anis Khan explains how trivago’s SRE teams act as internal consultants, helping product teams adopt reliable infrastructure through golden paths, shared Helm and Terraform tooling, reusable GitHub Actions workflows, and centralized observability. The post also covers knowledge sharing, guardrails, team self-sufficiency, and reducing operational silos.
Trivago explains how it built Price Stream, a scalable data product that aggregates billions of historical hotel prices across destinations, accommodations, dates, and locales. The post covers BigQuery grouping sets, Kafka-based delivery, deduplication, coverage constraints, and correcting outlier detection for non-normal price distributions.
trivago explains how its 80-person frontend organization works across product squads, QA, UX research, and product intelligence. The post covers large-scale A/B testing, automated QA with Selenium, Appium, Karate, and Playwright, and a successful viewport-based rendering optimization that improved performance and engagement.
Trivago explains how its Smart AI Search evolved from an early free-text search concept into a production feature using LLM-powered semantic search, traditional matching, filtering, and data preprocessing. The team also shares lessons on cloud partnerships, cross-functional development, and designing UX for users who struggle to express hotel preferences in free text.