XP.COM, LLC

Privacy-preserving data preparation for AI teams.

I help teams understand, transform, validate, and govern large legacy datasets before model evaluation, controlled sharing, or AI training work.

Discuss a scoped engagement View the full portfolio

1. Discover

Trace the data that actually matters.

Read application logic, schemas, and access rules. Map what is public, private, deleted, ambiguous, or otherwise excluded before a broad scan begins.

  • Visibility and access semantics
  • Schema, joins, lineage, and orphan checks
  • Scope rules for excluded records

2. Reduce risk

Separate detection from release.

Use deterministic redaction and screening, then route sensitive, ambiguous, or rights-risk material into quarantine for review. Automated signals are evidence, not clearance.

  • Direct-identifier and markup screening
  • NER and language signals for review queues
  • Media and sensitive-content risk controls

3. Prove the process

Leave useful audit receipts.

Preserve policy manifests, runbooks, counts, hashes, and release gates. Stakeholders can review what was done without distributing source records or row-level samples.

  • Aggregate-only metrics and reports
  • Reproducible validation records
  • Clear delivery and no-release gates

Founder-led case study

DDN archive data preparation

I applied this method to a mature community archive. The work covered approximately 8.96 million posts, 9.79 million postdata rows, 6.64 million chat rows, and 739,680 image rows. A strict internal forum validation covered 247,220 records.

The project used code-grounded visibility analysis, deterministic controls, integrity checks, NER-assisted review, sensitive-content quarantine, media risk profiling, and explicit release gates. It does not publish row-level data or represent an independent audit or legal opinion.

Request the engineering case-study summary

Start small

Begin with a scoped readiness assessment.

A first engagement can define source scope, identify major release risks, test the pipeline on a bounded corpus, and produce a prioritized technical plan.

Start a readiness discussion

Important limit

Engineering support, not legal advice.

I build and document technical controls. Counsel, rights holders, and accountable release owners must make legal, policy, and final release decisions.