Synthetic Data Specialist

A data fabrication engineer who transforms limited, biased, or privacy-restricted datasets into high-fidelity training resources that unblock ML pipelines — balancing statistical fidelity, formal privacy guarantees, and downstream model utility as three independent axes.

- The FABRICATE methodology — 8-pillar framework from source profiling through continuous improvement

- Generator selection playbooks for tabular, image, time-series, and text data with architecture tradeoff analysis

- Differential privacy integration with epsilon-delta budget allocation and formal privacy accounting

- Statistical validation protocols: marginal matching, joint distribution preservation, tail behavior replication

- Train-on-synthetic-test-on-real (TSTR) benchmarking to measure downstream model performance impact

- Privacy risk quantification using membership inference attacks and singling-out testing

- Automated synthetic data pipelines with schema ingestion, generation, validation, and versioned artifact publishing

- API design for self-service conditional generation enabling ML teams to request task-specific data slices

- Fidelity-privacy-utility triangle framework — measure all three axes independently, never hide tradeoffs

Drop into Claude, ChatGPT, Cursor, or any AI tool. Bring your real synthetic data problem — a privacy-frozen healthcare dataset, a rare-event imbalance, a regulatory compliance constraint, a model training bottleneck. It thinks like an engineer who's built generation pipelines that unblocked production ML systems.

Bundles or prompts related to ML data engineering and privacy-preserving machine learning.