TRUST & SAFETY · PLATFORM SAFEGUARDS

ABUSE INFRASTRUCTURE DISRUPTION · DETECTION ENGINEERING

ABOUT

Safeguards specialist who builds detection capability,
not just runs it

At Midjourney I co-built the platform protection function from zero: behavioral detection that finds coordinated networks in hours instead of weeks, graph attribution across fragmented identities, and enforcement strategy that turns findings into action across product, policy, and legal.

Enforcement at this scale is a precision discipline. A false positive is a real person locked out, so I build these systems to be sure before they are fast.

DETECTION TIME

AVG DETECTION TIME · COORDINATED ABUSE NETWORKS · 2023 TO 2026

NOTHING 30 DAYS UNDER 2 HOURS

RULE ENGINE + BEHAVIORAL SCORING + GRAPH CLUSTERING · ONE DETECTOR CORE, BUILT FROM ZERO · NETWORKS NOW IDENTIFIED BEFORE THEIR FIRST BILLING CYCLE COMPLETES

FEATURED WORK

Reseller Ecosystem DisruptionDisruption of third-party automation and reseller ecosystems behind hundreds of thousands of abusive accounts: behavioral fingerprinting, payment-signal analysis, and graph-based account clustering across fragmented identity systems, carried through to legal enforcement.CASE STUDYAbuse Investigation PlatformA solo-authored suite of analyst tools sharing one detector core across every surface: a rule engine, behavioral time-series scoring, and graph-based clustering that turn scattered signals into ranked, human-reviewed leads.SHIPPED · IN PRODUCTION USEGraph-Based Detection ResearchBenchmarking graph-neural-network detection models against production-shaped abuse topologies to test whether structure-aware approaches outperform rule-based detection as adversaries adapt.RESEARCH · IN PROGRESSEnforcement Workflow ApplicationA next-generation desktop application extending the shared detector core (Plate 02) end to end: moderation-signal ingestion, a pattern engine surfacing coordinated behavior, and verdict routing into automated enforcement with a human decision on every consequential action.IN DEVELOPMENT · BLUEPRINT

CAPABILITIES

Safeguards & Risk Detection

Abuse detectionBehavioral fingerprintingEntity resolutionGraph-based account clusteringRisk scoringPayment signal analysis

Investigation & Threat Analysis

Adversarial investigationsOSINTNetwork attributionThreat actor attributionVendor signal evaluationWorkflow automation

Software Engineering

Python (pandas)SQLSQLiteTypeScriptJavaScriptNode.jsElectronReactVueTailwindData pipelinesAgentic engineeringClaude CodeCursor

Systems & Operations

Detection frameworksEnforcement & disruption strategyAbuse economics modelingPolicy-to-product translationPre-launch abuse risk assessmentCross-functional communication (Engineering, Legal, Executive)

EXPERIENCE

MidjourneyFEB 2023 TO PRESENT

Platform Protection & Safeguards

  • Co-built safeguards and enforcement capability from nothing with a partner analyst: no formal mandate, data-constrained environment, function now embedded across product, policy, and legal.
  • Reduced average detection time for coordinated abuse networks from none, to roughly 30 days of manual work, to under 2 hours through behavioral analysis and custom investigative tooling.
  • Led disruption of third-party automation and reseller ecosystems responsible for hundreds of thousands of abusive accounts across three years of enforcement.
  • Designed behavioral detection combining payment signals, usage patterns, timing correlations, and prompt behavior; resolved fragmented identities into navigable networks via graph-based clustering without dedicated graph infrastructure.
  • Conducted adversarial investigations including OSINT attribution and proactive threat-actor research that informed enforcement and legal strategy.
  • Built evidence packets and investigation reports supporting outside counsel on enforcement matters.
  • Contributed to pre-launch abuse risk assessment and adversarial red-teaming for major model releases.

NOW

The abuse investigation platform (Plate 02) runs the daily investigative workflow in production. Two projects extend it. A next-generation enforcement workflow application (Plate 04) carries its detector core into automated verdict routing, with a human decision on every consequential action. A graph-detection research program (Plate 03) benchmarks structure-aware models against coordinated-abuse topologies, backed by a knowledge-graph index of 200+ papers narrowed to a candidate shortlist.

The open question driving both: whether structure-aware detection holds as adversaries adapt, or whether rule-based systems degrade faster than they can be maintained.

EDUCATION

Georgia State University, Robinson College of Business

B.B.A., Finance · Additional coursework in Computer Science, SQL, and databases