Drove structural transition for mini-app-centric expansion and improved search

Contributed to the AppsInToss initiative that enables users to access diverse mini-apps within the Toss app.

• Resolved limitations where the All Tab and Search could not represent mini-app sub-features

  • Redesigned the All Tab/Search data model and server architecture to support a mini-app and sub-feature hierarchy.

  • Established a flexible framework to expose and experiment with new mini-apps quickly.

• Implemented multilingual support across All Tab and Search data

  • Automated ingestion of dynamic and static multilingual data via a translation pipeline, replacing manual updates.

• Increased mini-app entry rate by improving the recommendation UI

  • Built a visualization of cumulative game plays in the All Tab.

  • Enabled mini-apps to appear in the All Tab recommendation module.

• Designed systems for scalable mini-app operations

  • Defined an integrated data flow across External Mini-app Console - Internal Service Manager - All Tab/Search Admin.

  • Systematized fragmented operational flows to improve efficiency and maintainability.

• Migrated from legacy keyword-based search to an ES-backed ML search service

  • Migrated from exact keyword matching to ElasticSearch-backed ML search by registering keywords in ES.

  • Streamed search data from Admin to ES via Kafka for reliable indexing.

  • Separated domain logic from the UI-facing layer to abstract and decouple the ML service.

Owned and experimented on key entry points: Home, All Tab, Notifications, and Search

• Removed structural constraints that blocked rapid iteration on high-traffic entry points

  • Decoupled and restructured Home, All Tab, and Notifications to enable experiment-ready services and targeting rules.

  • Introduced a Server-Driven UI (SDUI) experimentation framework to iterate UI quickly even with spec changes.

• Designed and operated user-centric experiments

  • Ran experiments on All Tab sort logic, pinned notification UI, and notification-blocking UX based on user feedback.

  • Improved conversion and dwell time through experiment-driven UI/UX changes.

• Eliminated performance and maintainability issues in legacy components

  • Upgraded to Spring Boot 3.2 and migrated to WebClient to reduce tech debt.

  • Isolated the All Tab service from legacy servers to remove cross-team deployment lifecycle conflicts.

Built and operated the Home spend-history refinement service for 18M MAU

• Standardized inconsistent exposure across diverse financial records

  • Maintained and extended a Merge-Refine pipeline to unify multiple data sources into a single refined spend history.

  • Onboarded new record types (card purchase cancellations, Coupang Pay, etc.) to improve accuracy.

• Enabled representation of additional metadata not supported by the legacy model

  • Designed and applied a SubTransaction structure to overcome model limitations.

  • Combined extra info (e.g., reward points, scheme info) with existing records for richer representation.

• Resolved server resource issues caused by excessive cache memory usage

  • Developed and adopted a schema-serializer library.

  • Reduced Redis memory by 60% and serialization CPU by 50% compared to MessagePack.

• Improved operational productivity with internal tooling and automation

  • Developed Timeline Admin V2.

  • Enabled merged-record inspection, TQL testing, and cache control directly in Admin.

  • Accelerated CS and ops response, reducing incident analysis time and improving response quality.

Modernized Hellobot backend architecture and delivered new features

• Addressed monolith and tech debt

  • Separated the high-traffic notifications server and DB from the legacy stack.

  • Removed a low-usability legacy admin and modularized components.

  • Extended data models for global expansion and added content management logic for global users.

  • Introduced test infrastructure to improve backend quality.

• Shipped multiple features aligned with business needs

  • Designed and implemented engagement features such as Random Box, Giftiel integration, and Scrap (bookmarking).

  • Iterated based on user feedback to enhance and harden features.

• Contributed to Hellobot globalization

  • Extended data structures to deliver content to global users.

  • Implemented logic to serve content only in allowed countries and languages.

• Automated operations for efficiency

  • Built an automated push-sending system for new skills to remove repetitive tasks and save ops resources.

• GPA: 3.90

• Lead Author

• https://arxiv.org/abs/2603.05302

• Development of near-infrared reactive multi-functional bead type hydrogel drug delivery system using Mesoporous Silica

• co-author

• https://www.dbpia.co.kr/journal/articleDetail?nodeId=NODE10724229