Agoda Feature Trails

Feature trailActive

AI-Assisted Software Development Workflow

Agoda is evolving its software development workflow to leverage AI coding assistants. This involves a shift from implementation to specification and verification, with a 'Grey Box' approach emphasizing clear intent, testable criteria, and evidence-based validation. The role of engineers is transforming from implementers to solution architects, focusing on system design and precise requirement communication. Specification becomes the primary interface, enhanced by AI for distillation and sparring. The OpsBot system has been enhanced with AI capabilities (OpsBot AI) to interpret natural language requests, gather context, and route them to the appropriate teams. Domain agents are being integrated into the workflow to perform automated investigations using approved tools and skills. An Agents Platform supports the creation and operation of these domain agents. Agent Eval is being introduced for evaluating agent changes and preventing regressions. Early success is seen in BI Product with significant coverage and reduced supporter time.

2 posts · 2026

Timeline

Feature trailActive

Price Cache with DragonflyDB

Agoda migrated its critical Price Cache from MSSQL to DragonflyDB to handle increasing data volumes and throughput. This involved a phased rollout, performance validation, and the implementation of a dual-cluster architecture for high availability and improved maintenance workflows. The new system offers better scalability, built-in expiry for volatile data, and reduced operational complexity compared to the previous sharded MSSQL setup. The migration included detailed parity checking and a gradual rollout strategy to ensure zero downtime. Post-migration, Agoda has implemented automated failover and optimized maintenance procedures using a dual-cluster setup.

1 post · 2026

Timeline

Feature trailActive

GitOps for Consul KV Management

Agoda implemented a GitOps workflow for managing HashiCorp Consul's Key-Value store. This involved creating a dedicated Consul KV Sync service and a Git repository mirroring Consul's structure. Changes are now managed via merge requests with CI validation against JSON schemas and CODEOWNERS reviews. The system automatically syncs Git changes to Consul across multiple data centers. This migration automated the onboarding of over 34,000 KVs and 100+ teams, eliminating configuration-related outages. Future work includes gradual rollout for global configurations and distributed ownership of CODEOWNERS.

1 post · 2026

Timeline

Feature trailActive

Flight Connector Workflow Simulator

Agoda developed a Workflow Simulator to test flight supplier connectors. This system models upstream callers, generates multi-step requests for specific business scenarios, preserves state across business processes, and asserts behavior across the entire workflow. It consists of a Scenario Builder, Test Runner, Workflow Executor (using a DAG to model interactions), and Assertions (endpoint and workflow-level). This allows for faster local testing and earlier detection of integration issues compared to traditional E2E testing.

1 post · 2026

Timeline

Feature trailActive

Design System Theming for White-Label Brands

Agoda developed a scalable design system capable of powering multiple travel brands. This system utilizes design tokens and a theming mechanism to allow for brand-specific visual customization without duplicating core component logic. The architecture involves core component libraries, base themes, and brand-specific theme packages that override design tokens. Runtime theming is managed via a Context Provider that injects CSS variables, enabling components to adapt their appearance based on the selected theme. This post details the implementation of design tokens and CSS variables for runtime theming, including the build process for theme NPM packages and testing tools like Storybook with a theme selector.

1 post · 2026

Timeline

Feature trailActive

Kotlin Coroutine Debugging and Performance Optimization

Agoda's Kotlin services have undergone significant improvements in coroutine management and debugging. This includes addressing issues with unbounded coroutines, optimizing serialization and deserialization processes, and resolving thread-safety bugs in open-source dependencies. The adoption of continuous profiling tools like Pyroscope has been instrumental in identifying and resolving these complex, layered problems, leading to substantial reductions in CPU and memory usage. This post details the technical implementation of the fixes for unnecessary deserialization, unbounded coroutines, and a thread-safety bug in Koin's ScopedInstanceFactory, resulting in a 5x reduction in resource usage.

1 post · 2026

Timeline

Feature trailActive

Multimodal Content System

Agoda built a Multimodal Content System to bridge the gap between siloed image and review data. This system uses a topic-based clustering approach to programmatically link visual content with textual feedback, providing users with a richer, contextual understanding of hotel properties. It involves enriching images and reviews with tags, mapping these to central topics, and serving curated content packages for low-latency retrieval. This post details the technical implementation of the image classification, review tag extraction, sentiment aggregation, and the overall data pipeline orchestration using PySpark and Kubeflow, with serving via Couchbase.

1 post · 2026

Timeline

Feature trailActive

Developer Onboarding and Documentation Hygiene

Agoda implemented a competitive 'Onboarding Olympics' event to rigorously test and improve the developer onboarding experience. This initiative directly addresses the gap between documentation (READMEs) and the actual system setup, revealing issues with outdated configurations, missing prerequisites, and unclear instructions. The event uses timed challenges where engineers attempt to set up unfamiliar repositories, with the owning teams observing live. This provides direct, actionable feedback, leading to improved documentation and setup processes.

1 post · 2026

Timeline