AWS officially featured GloZ’s translator matching platform built with Amazon OpenSearch Service and Amazon Bedrock, showcasing how hybrid search powers accurate matching across a pool of nearly 100,000 translators.
Published:
🎯 Featured on the AWS Official Tech Blog
On May 18, 2026, GloZ’s case study on building a natural language resume search platform was published on the AWS Official Tech Blog. The article marks the first in-depth case study from the AWS Korea Solutions Architects team analyzing GloZ’s multilingual hybrid search architecture for matching nearly 100,000 professional translators.
📌 Why GloZ Rebuilt Its Search Infrastructure
GloZ operates a global network of nearly 100,000 professional translators working across more than 70 languages. Every project match requires evaluating multiple layers of criteria:
- Language pairs (e.g. KO→EN, EN→JA)
- Domain expertise (gaming, medical, legal, etc.)
- CAT tool proficiency (Trados, MemoQ, etc.)
- Availability windows and response speed
The previous architecture — a combination of PostgreSQL and an in-house vector DB — required separate systems for keyword and semantic search. That setup created operational complexity while limiting search accuracy.
⚙️ Solution: Unified OpenSearch + Bedrock Architecture
GloZ consolidated its search infrastructure on Amazon OpenSearch Service. Key components include:
- Hybrid Search: Combining BM25 keyword matching with k-NN vector search (HNSW algorithm) in a single query
- Embedding: Amazon Bedrock’s Cohere Embed v4 for multilingual embeddings and Claude Haiku 4.5 for resume summarization and metadata normalization
- Nori Korean morphological analysis: Improved synonym handling and compound noun processing for Korean resumes
- ML Connector + Ingest Pipeline: Automatic embedding generation during indexing
📊 Key Outcomes
- Achieved nDCG@10 of 0.852 — approaching the target score of 0.90 for top-10 result quality
- Unified search across 30+ languages within a single index, including Korean, English, Japanese, Chinese, and Spanish
- Eliminated the operational burden of maintaining separate keyword and semantic search infrastructure
- Supported multiple resume formats including PDF, DOCX, and scanned images with OCR
💡 Key Insight: Data Quality Matters More Than Model Selection
One of the most important takeaways from the project was that data quality outweighed model selection. GloZ prioritized resume normalization through LLM-based metadata extraction, synonym mapping, and hallucination validation workflows, which ultimately became the biggest driver of search accuracy.
🔗 Source Article
The full technical breakdown is available on the AWS Official Tech Blog. By combining its 80,000+ translator network with AI-driven infrastructure, GloZ continues to help shape the technical standard for global content localization.