Korean LanguageSAMPLE
K-LINGO
Customer-Support Dialogue Summarization & Intent Model
Summarizes Korean call-center and chat transcripts and classifies customer intent
- Parameters
- 8B
- Intent F1
- 0.92
K-DATA MODEL LINEGlobal-standard data
Data becomes the model.
AI models trained on 14 fields of global-standard data, built by DATATREND.
When the data’s origin is clear, so is the model’s quality.
0Data fields
0+Datasets built SAMPLE
0Models
PIPELINE
Good models start
with good data
From data built at K-DATA to a finished model — one company owns the whole process.
SOURCE
K-DATA dataset construction
Labeled data built in-house across 14 fields
incl. AI-Hub construction track record
REFINE
Cleansing & cross-review
Duplicates and mislabels removed —
only training-ready data remains
TRAIN
Domain fine-tuning
Field-specific training with repeated
evaluation on standard benchmarks
DELIVER
Model delivery
Weights purchase · API subscription
· custom training
MODEL LINE-UP
Task-specific models,
trained per domain
Every model is trained narrow and deep — one task at a time. Follow a card’s link for the full model description.
RoboticsSAMPLE
K-ROBO
Human-Robot Collaborative Task Recognition in Home Environments
Recognizes task stages and actions in footage of human-robot collaboration at home
- Parameters
- 1.9B
- Action mAP
- 78.2
HealthcareSAMPLE
K-MEDI
Medical Consultation Summarization & Disease Classification
Summarizes clinical consultation records into structured form and classifies disease categories
- Parameters
- 4B
- Classification acc.
- 91.3%
Transport·LogisticsSAMPLE
K-LOGIS
Urban Freight Demand Forecasting & Route Optimization
Forecasts hourly freight demand from shipment history and optimizes delivery routes
- Parameters
- 2B
- Forecast MAPE
- 6.8%
FinanceSAMPLE
K-FIN
Financial Document Key-Information Extraction
Extracts and normalizes key fields from contracts, disclosures and other financial documents
- Parameters
- 4B
- Extraction F1
- 0.94
LawSAMPLE
K-LAW
Case-Law Retrieval & Legal Document Analysis
Issue-based case-law retrieval and clause-level analysis of legal documents
- Parameters
- 8B
- Case Recall@5
- 0.89
Video·ImageSAMPLE
K-SIGHT
Fall & Risk Detection for Seniors Living Alone
Detects falls and risk behaviors of seniors living alone from indoor video in real time
- Parameters
- 1.2B
- Risk F1
- 0.90
ManufacturingSAMPLE
K-FACTO
Battery Surface Defect Inspection
Detects and classifies fine defects on battery cell surfaces for quality inspection
- Parameters
- 0.9B
- Detection acc.
- 98.2%
EducationSAMPLE
K-ESSAY
Automated Essay Scoring & Feedback
Scores essay and descriptive answers against rubrics and generates feedback
- Parameters
- 4B
- Score agreement
- 0.88
Agro·FisheriesSAMPLE
K-AGRI
Honeybee Disease Diagnosis from Hive Images
Early diagnosis of honeybee disease signs from hive imagery
- Parameters
- 0.6B
- Classification acc.
- 93.5%
Disaster·SafetySAMPLE
K-SAFE
GPR-Based Road Subsurface Cavity Detection
Detects subsurface cavities and subsidence risk from ground-penetrating radar signals
- Parameters
- 1.5B
- Cavity F1
- 0.87
SportsSAMPLE
K-SPORT
Baseball Rule-Ruling Video Analysis
Recognizes key rule situations in broadcast footage to assist rulings
- Parameters
- 2B
- Ruling agreement
- 95.1%
Culture·TourismSAMPLE
K-TOON
Webtoon Cut & Scene Generation
Generates webtoon cut compositions and scene images from scripts
- Parameters
- 3B
- Scene consistency
- 4.3/5
IPSAMPLE
K-MARK
Similar Trademark Image Search
Searches and screens similar trademarks via image embeddings for IP protection
- Parameters
- 0.8B
- Recall@10
- 0.93
Korean LanguageSAMPLE
K-VOICE
Emotional & Style-Controllable Korean TTS
Synthesizes natural Korean speech with emotion and speech-style tags
- Parameters
- 1.1B
- Synthesis MOS
- 4.4
* Model names, parameters and benchmark figures are illustrative examples and will be replaced with actual specifications. The 15 sample models span all 14 data fields.
WHY K-AI MODELS
A data company first,
a model company second
DATATREND built data long before it built models. Knowing the data is what makes the models different.
Data with a clear origin
We build and sell datasets across 14 fields at K-DATA. These are not models trained on data of unknown origin.
Source data built to global standards
Not repackaged third-party data — collected, labeled and reviewed to international standards, so performance on real tasks is tangibly different.
License-clear training data
Only data cleared for copyright and privacy goes into training. Ship it in your service without legal risk hanging over you.
Models that grow with the data
Every time a new dataset is built, the models are updated with it — an update cycle only a data company can run.
CONTACT
Adopting an AI model?
Talk to us today
Tell us your use case and domain, and we’ll help design the right model and adoption path together.
Buy datasets at the K-DATA store