ReSeorch.AI리서치에이아이ReSeorch.AI
Seo Gyeongmin · Registered sole proprietor · Korean Text · Audio & TTS
I evaluate Korean text and audio/TTS outputs as an independent contractor. My mobile software QA background informs how I apply requirements, compare outputs, identify ambiguous cases, and document the reasons for my judgments.
In personal projects, I use AI assistance to develop supporting tools while keeping my own evaluations distinct from technical checks and research findings.
Voice evaluation and QA draw on work experience. Workflow design and Korean context are areas of independent project work and research.
AI evaluation work
Compare speaking intervals and response timing
Software QA work
Reproduce the problem and narrow the cause
Independent project · workflow design
Separate AI tasks from human decisions
Independent research · Korean context
Use context to narrow an interpretation
Korean Text · Audio & TTS
Outlier · Independent Contractor
I review Korean text and audio/TTS outputs against provided requirements and evaluation rubrics. My work includes comparing outputs, identifying errors and ambiguous cases, and documenting the reasons for my judgments.
A personal project, separate from contract work.
Mobile applications · Telecom services
Test & C Inc. · Full-time employee
Performed defined functional, compatibility, and regression tests for mobile applications and related telecom services. Reproduced and documented defects, recorded test conditions, and verified fixes across mobile and wearable environments.
A personal project, separate from employer work.
Bachelor's studies in culture and liberal arts
Public personal projects, separate from my contract work.
June 2026 · Personal evaluation project
I reviewed an AI response that interpreted a Gyeongsang dialect passage in standard Korean. I considered the response usable while recommending a less categorical explanation of one time expression. The same response already included a broader interpretation, so my note does not treat the whole answer as a mistranslation.
The translated sentence used “the day before yesterday”; an explanation in the same response also allowed “a few days ago.” I recorded factual accuracy as 3/4 and Korean context fit as 4/4. These are my original judgments from a personal evaluation project, not new scores.
This is one of 10 prompt responses I evaluated against 8 criteria, recording scores, defects, reasons, and suggestions. I performed the evaluation; AI assisted with the templates.
Read the original response and my evaluation (Korean)June 2026 · Personal QA project / October 2026 · Site revision
A personal website review and a basic Playwright suite for key elements, link attributes, and visibility at a narrow viewport. The June test record is kept separate from current verification.
For the October revision, I defined requirements and reviewed content and reading flow. AI assisted with implementation and checks for responsive layouts, navigation, and keyboard behavior.
Web QA case study (Korean)Notes on input handling, preserving source material, and reviewing exceptional states.
The existing portfolio evidence describes review of source preservation and input/save workflows.
Project review notes
The current operational and publication status of the real project has not been verified.
Explore the project and demo (Korean)Reviewed input structures, exception states, source preservation, and workflows in a personal spreadsheet-based operations tool. Implementation was AI-assisted.
Project review notes
Completion of the real ledger and settlement automation has not been verified.
Explore the project and demo (Korean)Apply requirements, compare outputs, and explain differences with supporting reasons.
Related work (Korean)Review naturalness, tone, and context; avoid treating insufficient evidence as a firm conclusion.
Related work (Korean)Record conditions, steps, and observations so that an issue can be reviewed again.
Related work (Korean)Basic regression tests for a personal website. With AI assistance, I expanded checks for responsive layouts, navigation, and keyboard behavior.
Basic queries and aggregation, including JOIN, NULL, and subquery concepts.
Basic scripting and data review. In personal projects, I use AI-assisted tools for checking inputs and records and organizing results.
Manage source files, outputs, and change history for personal projects.
In personal projects, I review requirements, inputs, and outcomes. AI assists with implementation and testing; that support is separate from my direct evaluation work.
Personal evaluation projects developed with AI assistance. Human judgments, technical checks, and research findings are recorded separately. No hypothesis findings are available yet.
Does more reasoning improve the fit of a Korean response to its context?
Limited live connectivity checked · Context review incorporatedA personal evaluation project on Korean conversational appropriateness. A limited live connectivity test is complete, and human review of case context has been incorporated. Final human response ratings and comparative findings are not yet established.
My contribution. I review necessary context, unsupported assumptions, exceptions, and conditions for withholding judgment. Candidate drafting and tool implementation use AI assistance.
Limits. Design feedback is distinct from final response ratings. No claim about broad cultural understanding or comparative performance has been established.
Content checked:
Project note in KoreanHow well do predicted response behaviors match observations?
Collection and judgment entry complete · Validation incompleteA personal project comparing predicted model behavior with observed responses. Response collection and human judgment entry are complete. Some entries were made in a batch after reading AI-compressed summaries; validation against individual original responses and predictive analysis are not complete.
My contribution. I review the research question and scope and record judgments and reasons after reading responses or AI-compressed summaries. Tool implementation and technical checks use AI assistance.
Limits. Entry completion is distinct from independent review of every original response. No predictive performance finding has been established; this work does not establish consciousness, intent, deception, or general self-understanding.
Content checked:
Project note in KoreanDo explanations stay consistent when evidence and conditions change?
Execution tooling tested offline · Before response collectionA personal project examining whether AI explanations remain consistent with the evidence and conditions provided in fictional scenarios. Pre-execution tooling has been tested offline; model responses have not yet been collected.
My contribution. I define and review the research question, scope, and conditions for proceeding. Design documentation, implementation, and tests use AI assistance.
Limits. Offline tool checks do not establish model performance or findings about consciousness, rights, or the validity of human judgments.
Content checked:
Project note in KoreanResearch summaries v2.1.