DTS is CUBIG's AI-ready data transformation engine. It generates DP-protected datasets using differential privacy to fix class imbalance, fill coverage gaps, expand training data, and replace restricted or non-accessible data. DTS can be deployed on its own for data transformation work, and operates as a core capability of the Syntitan platform.
DTS · AI-ready data transformation engine
Rebuild unusable data
Rebuild unusable data
into AI-ready datasets.
Most enterprise data isn't AI-ready. DTS rebuilds restricted, imbalanced, or incomplete data into an AI-ready dataset you can actually use.
It replaces restricted data with privacy-safe substitutes, rebalances skewed datasets by generating additional data, and fills gaps in what the data covers with new AI-ready data.
Data problems
Three data problems. One engine.
Data that can't be shared, can't be used, or can't be accessed. DTS resolves all three.
Capability
Privacy-safe synthetic data, as a capability.
Synthetic data is one capability inside DTS, not its identity. DTS uses it, with differential privacy underneath, to expand coverage and repair imbalance when real data can't be used.
Differential Privacy
A formal privacy bound, by design.
What differential privacy means
Differential privacy (DP) is a mathematical framework that bounds how much any single individual's data can influence the synthetic output. Individuals cannot be re-identified, no matter what outside information someone combines it with.
DTS applies DP during the generation process itself, not as a post-processing anonymization step. The privacy property is structural, not dependent on masking or field removal, a provable bound, not best-effort masking. This formal privacy bound is backed by our own research (MPGAN, BMVC 2022) and a registered patent.
The bound
The chance of identifying any individual from the synthetic dataset is capped by a defined value, epsilon (ε), regardless of outside knowledge.
How DTS generates synthetic data
01
Statistical profiling
DTS analyzes the real dataset's statistical properties (distributions, correlations, and other statistical patterns) without storing raw records.
02
DP noise injection
Calibrated noise is injected into the statistical model according to DP bounds, so individual data points become mathematically unidentifiable.
03
Synthetic generation
New records are sampled from the DP-protected model. Output is statistically representative but contains no real personal information.
04
Fidelity validation
Generated data is validated against the original distribution. Quality and utility metrics confirm suitability for training and validation use.
Deployment
Start with DTS, grow into Syntitan.
Mode A · Direct
DTS on its own
DTS is a core capability of Syntitan you can start with directly, against your own data sources. It fixes AI training-data quality, generating what's missing at scale without touching real data.
- Fix class imbalance: generate more examples of rare classes with distribution fidelity
- Augment sparse datasets to production-grade volume
- Generate edge cases and rare-event samples
Mode B · Integrated
DTS + Syntitan
When compliance blocks data from reaching models, DTS runs inside Syntitan to generate privacy-safe replacements. DTS makes the data. Syntitan versions and tracks it.
- Replace GDPR, PIPA, HIPAA-restricted data: the original data never leaves your environment
- Syntitan versions the synthetic dataset and binds it to a Release State
- Syntitan's change log tracks it from data generation through the AI run
In production
Finance · IBK Industrial Bank
97.6% fraud-detection accuracy (AI model) · 79 patterns → 1,000 records
Fraud and transaction patterns expanded into DP-safe synthetic records. PIPA-compliant, with zero real customer data exported.
Finance · Kyobo Life Insurance
F1 0.92 churn model · 277,249 synthetic records
A 6-month data-retention policy had blocked Kyobo's churn AI. DTS rebuilt DP-safe records from historical data, legally usable after deletion.
Marketing / Sales
90% time reduction · 70% cost saving on trend research
Annual consumer-trend surveys replaced with AI persona agents trained on synthetic behavioral data. Insights in 1 to 2 days instead of a month.
Defense · Ministry of National Defense
Zero data exports · classified imagery → AI-ready
Deployed on-premise in an air-gapped classified environment. Classified data became AI-ready synthetic datasets within clearance.
Comparison
DTS vs. other approaches to restricted data.
| Capability | DTS | Masking / Anonymization | Data Sampling | Manual Labeling |
|---|---|---|---|---|
| Privacy bound | ✓ Formal DP bound (ε) | △ Re-identification risk remains | ✗ None | ✗ |
| Coverage expansion | ✓ Generate at any scale | ✗ Can't create new data | △ Bounded by real data volume | △ Expensive & slow |
| Rare-class augmentation | ✓ Targeted generation | ✗ | ✗ Can't create rare events | △ Very high cost |
| Distribution fidelity | ✓ Validated against real stats | △ Distorted by masking | △ Sampling-bias risk | △ Annotator variance |
| Cross-border / external use | ✓ No real data transferred | ✗ Residual risk | ✗ | ✗ |
| Syntitan integration | ✓ Native versioning & binding | ✗ | ✗ | ✗ |
When to use
Five signals your data is blocking AI.
Enterprise AI projects stall when data conditions prevent training, validation, or safe deployment. DTS was built for these situations.
Restricted Data
Data exists but compliance blocks AI access.
GDPR, PIPA, HIPAA, or internal retention policies prevent the data from reaching models.
Unusable Data
Imbalanced datasets or coverage gaps distort model behavior.
Rare classes underrepresented, fraud patterns too sparse, edge cases absent from training.
Unusable Data
Retention policies delete what AI needs.
Historical data was deleted per retention policy, so the patterns that trained the previous model no longer exist.
Restricted Data
Sensitive records can't leave your environment.
Classified, patient, or customer data cannot be exported for AI training, even internally.
Unusable Data
Training-data volume is too low for reliable AI.
The original dataset is too small to train a robust model, and collecting more takes months.
Outcome
In each case, DTS turns data that is restricted or unusable into an AI-ready dataset, without exposing real records.
See if DTS fits your data
Proof
Proven in production.
+30pp
F1-Score Lift
58.55% → 88.55%
−90%
Time to Deploy
4 weeks → 1 day
97.6%
Fraud-Detection Accuracy (AI model)
IBK Industrial Bank
277K+
Synthetic Records
Kyobo Life Insurance
Listed as a Representative Vendor in Gartner®, Emerging Tech: Provider Differentiation Strategy–Trends for Hyper-Synthetic Data (2025).Gartner does not endorse any vendor, product or service depicted in its research publications. GARTNER is a registered trademark of Gartner, Inc. and/or its affiliates.
FAQ
Frequently asked questions
Differential privacy (DP) is a mathematical framework that puts a hard bound on how much any single person's data can influence the output. This keeps re-identification risk low, no matter what outside information someone combines it with. DTS applies DP during generation, so datasets stay statistically representative while containing no real personal records.
Yes. DTS can be deployed on its own for transformation workloads. As part of Syntitan, its datasets are versioned and bound to Release States.
Three categories. First, restricted data that privacy or compliance rules keep from being shared. Second, data with coverage gaps or class imbalance that makes models unreliable. Third, data that exists but cannot reach training pipelines.
Original data stays inside the client environment. DTS analyzes statistical properties in place, and only the DP-protected synthetic output moves on. No raw records are transferred outside. This makes the architecture suitable for environments where data cannot move: classified, regulated, or isolated networks.
DTS is the transformation engine; Syntitan is the platform it powers. Syntitan performs data-quality refinement as part of execution stability and can use a subset of DTS capabilities when DP-protected synthetic data is needed, while DTS is the platform's full AI-ready data transformation engine, which can also be deployed on its own.
Restricted data. Usable AI.
DTS rebuilds the data your AI can't use today into datasets it can train on tomorrow. GS Certified. KISA approved.