shape

UClone - Optimizing Large-Scale Data Search with OpenSearch

Overview

Resolving search performance bottlenecks from growing conversation data and improving AI response quality

Tech Stack

  • Go

    Go

  • MySQL

    MySQL

  • PostgreSQL

    PostgreSQL

  • Redis

    Redis

  • OpenSearch

    OpenSearch

  • Docker

    Docker

  • GCP

    GCP

Team

Software Engineer Intern

Period

2026.01 ~ 2026.03

Details

  1. Overview
    • UClone is an LLM-based social platform that creates AI clones reflecting a user's persona and uses them to support task delegation and social collaboration.
    • Working remotely for a Silicon Valley startup, I was responsible for the service's backend and frontend development as well as cloud infrastructure operations.
  2. Background
    • As conversation data grew, the existing DB-based search hit a performance bottleneck.
    • In addition, the limited context that could be injected into the LLM constrained our ability to improve AI response quality.
  3. Solution
    • Built an OpenSearch cluster and migrated the existing DB search to it, improving search response times.
    • Designed an embedding pipeline for conversation data and introduced vector indexing, enabling more context to be retrieved and injected.
  4. Additional Contributions
    • Developed full-stack (web and app) features, including a Go server, and operated the production servers.
    • Built a GKE-based load balancer and integrated and optimized LLM APIs to reduce fixed costs.
    • Gained experience with agile, daily-standup-driven development while working remotely with a Silicon Valley company.