18 years enterprise operations → AI systems

Justin Erwin

AI Product Manager & Technical Builder

I build AI systems that have to work in the real world.

I spent 18 years leading high-volume resort operations, then moved into product to build the tools I always wished existed on the floor. Today I design AI systems for complex workflows, real operational constraints, and measurable business outcomes.

View workContact
Reztrix · system snapshot
Grounded AI for frontline execution
293 REST endpoints across 33 routers
69 public tables with 339 RLS policies
50 synthetic scenarios. 84% pass rate, up from 42% baseline
FastAPIpgvectorLLM-as-JudgeSSEHITL
293
endpoints
69
tables
42→84%
eval pass rate
01Proof Strip

Operational Scale. Product Impact.

300%
REVENUE LIFT

Led an IoT smoke-detection pilot across 1,000 rooms, contributing to a 300% revenue lift, $500K+ in compliance revenue, and fewer guest complaints.

42% to 84%
EVAL PASS RATE

Built a custom LLM-as-judge framework across 50 synthetic scenarios, improving pass rate from 42% to 84% through iterative root-cause analysis.

02Tier 1 · Live Flagships

Three projects I'd show you first.

AI OPERATIONS PLATFORM · SYSTEM PROTOTYPE
Reztrix: Frontline Ops Intelligence

AI decision-support system for frontline operations in complex physical environments. Built to connect fragmented property workflows with safer AI-assisted action, including evaluation pipelines, fallback logic, and human approval where the cost of error is high.

FastAPIpgvectorRAGHuman-in-the-Loop
DATA PRODUCT · LIVE TOOL
BurgReport: Fine Wine Pricing Intelligence

Live revenue intelligence tool that aggregates and normalizes external market data into fast, usable pricing insight. Built around API ingestion, messy-data standardization, and sub-3-second response performance.

Next.jsAPI IntegrationData NormalizationVercel
KNOWLEDGE SYSTEM · LIVE PROTOTYPE
Spatial Limen: Structural Inference Engine

Evidence-grounded structural inference system for comparing complex information while preserving provenance, uncertainty, and reviewer control. Designed to surface deeper relationships beyond standard search or chatbot interfaces.

Next.jsThree.jspgvectorProvenance
03Tier 2 · Enterprise Strategy

Product strategy for regulated workflows.

PRODUCT STRATEGY · RESEARCH ARTIFACT
Regulated Transfer Exception Workflow

Research-backed product strategy for AI-assisted fraud review, structured evidence assembly, and audit trails in regulated financial transfers. Designed around strict timing constraints, human-in-the-loop review, and institution-controlled decisions.

ACATSEvidence WorkflowAudit TrailHITL
04Technical Credibility

The stack I actually use.

MODELS & EVALS
Claude SonnetLLM-as-JudgeSynthetic scenariosStructured output
RETRIEVAL & GROUNDING
RAGpgvectorPolicy embeddingsSupabase
BACKEND & DATA
PythonFastAPIPostgreSQLREST APIs
INFRASTRUCTURE & SECURITY
RLS policiesEdge Functionspg_cronSSE streamingHuman-in-the-loop
EDUCATION & CREDENTIALS
MBA, Kelley School of BusinessStanford GenAIAWS Certified AI PractitionerAWS Certified Cloud PractitionerPSPO I

Building something that has to work?

Senior PM and PM roles in AI-forward SaaS and hospitality tech. Remote US or Las Vegas.

Email JustinLinkedIn