Toss implemented agent task continuity in network-segmented environments using bubblewrap isolation and PV-based state storage.
Toss TASTE boosted recommendation performance by 35.1% using multitask learning and applied it to other services.
Toss significantly reduced LLM invocation costs by applying Selective Neural Approximation (SNAP).
Toss converted log definition tasks into AI skills to achieve consistent specification management based on domain context.
In Toss Securities Oracle DB, repeated commit waits of 0.1 seconds or more despite normal average CPU were caused by NIC interrupts concentrated on one socket, and most of it improved after distributing interrupts across two sockets.
By switching to domain-specific clusters and dbt, the advertising domain's DAG code was reduced from 2,794 lines to 100 lines.
Delegates syntax transformation and bundling to the host while preserving original structure and supporting declaration sourcemaps
Support for require(esm) since Node.js 22.12.0 resolves CJS compatibility issues and reduces maintenance overhead
Resolved server replacement and memory leak issues at the tenant level alongside Apache Spark Operator.
Offering a 10% annual interest rate for daily savings of 5,000 KRW, the product has established itself as an entry-level option for savings and deposits.
Automated model invocation, security, and cost management through the GenAI Portal and Gateway.
Performance improved on the same resources by reducing the remote DRAM access ratio from 50% to 0.8%.
Reducing the number of candidates by half increased orders per user by 10.13% while decreasing CTR by 3.71%
AI generated 101 drafts, which were refined to finalize 9 character designs.
Toss Bank introduced AI consultation for jeonse loans, saving 300 hours of consultation time annually.
Toss Bank separated its asset management report into a floating card, improving click-through and conversion rates by 3x.
Toss Doctor V2 combines LLMs and Appium MCP to automate everything from test code generation to failure recovery, extending to regression testing via Toss Checker.
A Toss Bank designer shared how user interviews and firsthand experience shifted the team's core value for international remittances from 'low fees' to 'alleviating anxiety'.
Toss Bank has unveiled the UX principles behind its 'Locked Piggy Bank,' designed to curb impulsive spending among teenagers.
Toss analyzed the discrepancy between public benchmarks and actual service performance, establishing Toss Benchmark as a domain-specific evaluation standard.
Toss Securities shared its experience of overcoming the limitations of GPU-awareness by building a GPU-native cluster equipped with validation and resource control capabilities.
Toss Bank decided to retain Pages Router after experimental adoption of App Router revealed that operational costs exceeded the benefits.
Toss Bank developed 'Stylepack' to resolve Coding Agent team rule violations by injecting rules immediately after file creation and at session termination.
Toss has built a knowledge management system through its internal documentation platform 'todoc', which supports AI integration and automated validity management.