NeMo Data Designer vs Synth Studio

NeMo Data Designer is a compound AI system for generating synthetic data. Users configure dataset columns and models, preview sample records, then create a dataset. Its framework manages batching and parallelism, checks generated data, and supports statistical diversity, field correlations, and reproducible workflows. It handles tabular data, text, code, structured JSON, images, and embeddings, with CSV, JSON, JSONL, and Parquet output formats. The system is available as an open-source Python library or as a NeMo Microservice with REST API methods for previewing and creating datasets. Documented LLM endpoints include NVIDIA API Catalog, OpenAI, Azure OpenAI, Anthropic, vLLM, TGI, TensorRT-LLM, and OpenAI-compatible servers. The microservice also connects with NeMo Inference Microservices, NeMo Customizer, and NeMo Evaluator. Data Designer does not host models, so users need to supply LLM endpoints and inference capacity. The hosted free trial requires an API key; jobs default to 10 records and allow up to 100 per job. The default SECURE Jinja renderer includes sandboxing and output safeguards.

Synth Studio generates synthetic tabular datasets with mathematical privacy guarantees. It accepts CSV and JSON files and detects their schemas automatically. Generation options include CTGAN, TVAE, GaussianCopula and schema-based methods. Differential privacy settings can be configured with epsilon and delta, using RDP accounting. Quality tools assess statistical similarity, machine-learning utility and privacy leakage, while automatic PII detection identifies fields such as names, emails and SSNs before generation. The product provides compliance reporting for HIPAA, GDPR, CCPA and SOC-2, alongside audit logs and model cards. Synth Studio is open source and can be self-hosted or used through its cloud offering. Local installation documentation lists Windows 10+, macOS 10.15+ and Linux. The free plan is free for up to 1 million rows per generation under an MIT License. REST API integration is available, but direct API access currently requires a session cookie or trusted proxy headers; API keys are upcoming. Documented API limits include 1,000 authenticated requests per minute and five concurrent generation jobs.