About Secludy
Move faster with sensitive data
Secludy helps enterprises train, test, and evaluate AI on realistic synthetic data without exposing raw PII, PHI, or proprietary customer data.
We turn restricted datasets into privacy-guaranteed synthetic data that AI teams can use and governance teams can approve.
Our mission
Secludy helps enterprises ship AI faster by making sensitive data safe to use. The highest-value data is often the most sensitive. We close that gap.
The problem
The best data is the data teams cannot touch
AI teams are told to ship. Their strongest data sits behind privacy, legal, security, compliance, and customer contracts.
Masking destroys the signal
Redacted and masked data loses the patterns models need most. What is left is safe and far less useful.
Approvals stall the work
Teams wait weeks on review while reviewers have no safe artifact to inspect and sign off on.
Models ship underpowered
Without the right data, models underperform. The business pays for a privacy gap dressed up as an AI gap.
The outcome
Privacy becomes an AI accelerator
Secludy turns a blocked workflow into an approved one. Teams build on real signal while raw data stays protected.
Move faster with AI
Ship models on the data you actually need instead of waiting on the data you are allowed to touch.
Unlock blocked data
Turn restricted datasets into a usable synthetic artifact your AI team can build on today.
Reduce leakage risk
Cut the exposure of PII, PHI, and proprietary customer data before it reaches a model.
Shorten review cycles
Give privacy, legal, security, and compliance teams a clear path to yes.
Hand governance real evidence
Deliver audit-ready reports with privacy parameters, lineage, and reproducible results.
Keep raw data in-house
Run self-hosted so your sensitive data never leaves your environment.
What we build
One safe dataset for the whole AI pipeline
GraphReplica produces synthetic data your team can build on and your reviewers can trust. Every capability maps to a buyer outcome.
Privacy-guaranteed synthetic data
Realistic synthetic data for tabular records and unstructured text, so teams build on the signal that matters.
Full AI workflow coverage
Data for training, testing, fine-tuning, and evaluation, so one safe dataset serves the whole pipeline.
Leakage testing
Membership inference and canary checks that prove sensitive values do not survive into the output.
Utility benchmarking
Side-by-side scoring against real holdout data, so teams trust the synthetic set before they train on it.
Audit-ready reports
Privacy parameters, data lineage, and reproducibility in one record governance teams can sign off on.
Rare-case enrichment
More examples of rare events like fraud, so detection models learn from cases real data barely contains.
What we believe
Privacy should ship AI, not stop it
Privacy should help companies ship AI safely, not stop them.
AI teams should never have to choose between speed and responsible data use.
Privacy belongs in the workflow from the first step, not bolted on at the end.
Legal, security, privacy, and AI teams need shared evidence so they can decide faster.
Our story
Built to get the right data approved
Founded in San Francisco in 2024.
The founding insight
The founders met through Y Combinator co-founder matching. They kept seeing the same blocker. The hardest part of getting AI into production is often not the model. It is getting approval to use the right data.
Teams need a safe data artifact they can inspect, test, approve, and use. That belief shapes everything we build.
From fine-tuning to a usable dataset
Secludy first explored privacy-preserving fine-tuning. Customer conversations pointed somewhere clearer. Teams wanted a usable, inspectable, approvable dataset they could hand to both AI and governance.
So we pivoted to privacy-guaranteed synthetic data, with differential privacy as the mathematical foundation underneath the guarantee.
The team
Meet the founders
Secludy is built by a team that has lived both sides of the problem, enterprise privacy risk and deep ML infrastructure.
Ben Cerchio
Co-founder and CEO
Ben has spent his career in privacy, risk, InfoSec, and compliance. He worked on product privacy at TikTok, technical privacy and InfoSec compliance at PayPal, and third-party risk at J.P. Morgan Chase.
He saw how privacy, risk, and compliance requirements shape what teams can safely build. His experience made one thing clear. AI teams cannot move fast if the data they need is blocked.
Mingze He
Co-founder and CTO
Mingze holds a PhD in Computational Biology and Bioinformatics with a focus on machine learning. His research is published in top journals including Nature, with more than 2,000 citations.
He has led enterprise ML and AI teams at scale, with roles as AI and ML lead at Williams-Sonoma, ML scientist at Stanford Research Institute, and machine-learning visiting scholar at UC Berkeley.
Who we serve
Built for teams with valuable, restricted data
We start where the need is most urgent and expand from there. Any company with proprietary customer data needs a safer way to use it for AI.
Fintech and financial services
Our first focus, where the pressure to use customer data and the cost of exposing it are both highest.
Healthcare and life sciences
Teams that need realistic patient and research data without moving PHI into a model.
Consumer tech and marketplaces
Companies building personalization and trust models on data they cannot pass around freely.
B2B SaaS
Product teams that hold proprietary customer data under contract and still need to train on it.
Trust and deployment
Evidence on every run, raw data that never leaves
Secludy is designed for the teams that have to say yes. The reviewers get proof, and the sensitive data stays put.
Self-hosted by design
Deploy on-prem or in your VPC so raw data stays inside your environment from end to end.
Audit-ready evidence
Every run produces reports with privacy parameters and lineage your reviewers can verify.
Leakage testing built in
We test the output for memorized sensitive values, then show governance teams the result.
Built for regulated work
Built to meet GDPR, CCPA, and HIPAA requirements, with privacy controls you can configure and inspect.
Bring a dataset. We will show you the safe version.
Book a demo and we will generate a privacy-guaranteed synthetic replica on a sample of yours. First safe dataset in under a week.