KT and Amorepacific Launch Lemon Research Assistant
KT and Amorepacific launched Lemon on July 28, an internal generative-AI assistant built on a reorganized data platform spanning more than 70 years of cosmetics R&D records. The companies say researchers can query ingredients, formulations, experiments, and reports in natural language; public coverage does not disclose the model, retrieval architecture, evaluation results, or independent performance testing.
KT and Amorepacific launched Lemon, an internal generative-AI assistant for cosmetics researchers, on July 28, 2026. The tool sits on a broader data platform that reorganizes more than 70 years of Amorepacific research records for natural-language search and future AI services.
The data project behind Lemon
The Korea Times reports that KT completed a project called Data Highway after winning the work in December. The companies standardized structured and unstructured records from Amorepacific's research organization into what they call "AI Ready Data." The reported collection includes millions of records covering ingredients, formulations, experimental results, research reports, patents, and papers.
Lemon stands for Lab Efficiency Mode ON. Researchers can ask questions in natural language and receive answers assembled from several categories of internal research material. KT says work that previously required searches across multiple systems over days or weeks can now take about five minutes. That is a company-reported workflow claim; none of the retrieved coverage provides an independent timing test or a task-by-task accuracy result.
The platform is also intended to serve as a shared data layer for future Amorepacific AI services and agents. The companies say they plan to keep expanding the organized research corpus and extend AI use across research and development.
What remains unreported
The launch coverage does not identify the underlying language model, embedding or retrieval stack, ranking method, citation interface, access-control design, or evaluation framework. It also does not say how Lemon handles conflicting experiments, superseded formulations, unpublished records, or questions whose supporting evidence is incomplete.
Those omissions matter in scientific R&D. A fast answer is useful only if researchers can trace it to the correct experiment, version, and context. Teams building similar systems would need document-level permissions, provenance for every answer, retrieval tests by source type, clear abstention behavior, and monitoring for answers that combine incompatible records.
The most transferable lesson is the sequencing: standardize the research data before expanding the assistant. Lemon is a concrete internal deployment, but the public evidence supports the existence and intended workflow of the platform, not a conclusion about scientific accuracy or product-development gains.
Key Points
- 1Lemon provides natural-language access to a reorganized internal corpus spanning more than 70 years of Amorepacific cosmetics R&D records.
- 2KT says some cross-system research searches can fall from days or weeks to about five minutes, but no independent timing or accuracy evaluation was retrieved.
- 3Public coverage does not disclose the model, retrieval architecture, provenance interface, access controls, or evaluation framework.
Scoring Rationale
Lemon is a concrete vertical AI deployment built on a substantial enterprise data-reorganization effort, but public evidence lacks architecture, evaluation, governance, and independently tested performance details.
Sources
Public references used for this report.
Practice with real Ad Tech data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
See all Ad Tech problems
