Ray Swan_

Resume

01 · Summary

Architect, builder, product lead

Architecting deterministic safety frameworks and multi-agent runtime environments for the past three years has redefined how I approach production artificial intelligence. Rather than treating large language models as black-box oracles, I engineer the rigorous structural scaffolding required to make them reliable enterprise assets—focusing on dual-kernel proposal-verification pipelines, hyperbolic graph neural networks, and fault-tolerant cloud orchestration.

My technical foundation is built on full-stack mastery across React, TypeScript, Python, and cloud infrastructure on AWS and GCP, allowing me to bridge the gap between low-level algorithmic research and scalable software architecture. But what makes me an exceptional candidate for an AI/ML Systems Architect role is the operational discipline I bring from the real world. Managing enterprise-grade software integrations alongside high-throughput physical operations—from full-stack platform launches to directing complex school nutrition logistics—has trained me to design systems that are not only theoretically sound under pressure, but structurally resilient, regulatorily compliant, and built for humans to actually use.

02 · Expertise

Architecture & engineering

AI Platforms & GenAI

Azure ML; Azure Databricks; Microsoft Foundry; MLflow; RAG; vector retrieval; grounded assistants; agentic systems; MCP-oriented tool boundaries; model evaluation

MLOps / LLMOps

Local-to-cloud lifecycle; champion/challenger; model registry; promotion/retain gates; drift monitoring; retraining; rollback-ready versioning; observability-ready contracts

Cloud & IaC

Terraform; GitHub Actions; Entra ID/OIDC; RBAC; Docker; Azure; GCP/Vertex AI; AWS/SageMaker; APIs; microservices; PostgreSQL

Governance & Security

Digest-bound saved plans; least-privilege identity separation; immutable evidence/provenance; state-machine guardrails; deterministic validation; bounded permissions

Azure MLAzure DatabricksMicrosoft FoundryMLflowTerraformGitHub ActionsEntra / OIDCMLOps / LLMOpsRAG / vector retrievalAgents / MCPModel evaluationDrift & retrainPostgreSQLProduct strategy
03 · Selected architecture

Azure GenAI/ML Ops Factory

Architect & Hands-on Engineer · 2026
  • Re-engineered Microsoft's Azure MLOps v2 Accelerator into a factory that generates modular, self-contained Azure ML project repos with governed infrastructure and configurable behavior.
  • Terraform-managed Azure ML: keyless storage, Key Vault, Log Analytics, scale-to-zero compute, Entra/OIDC GitHub CI, and evidence-bound plan/apply — destructive drift blocked before apply.
  • Proved a local-first lifecycle — shared Python logic locally and as a four-stage Azure ML pipeline (prepare → train → evaluate → register) — live MLflow run registered in Dev.
  • Built a Databricks retrieval proof on governed Gold data with three Vector Search indexes (notes, tickets, playbooks); filtered retrieval returned the correct evidence first.
Full case study — AI/ML Ops Factory ↗
04 · Additional systems
Eidetix Bio Research · 2025–present

Tokyo Eye — Scientific AI / Drug Discovery

  • Architected a governed scientific-AI platform using hyperbolic GNN and atom-level Transformer/Mixture-of-Experts architectures for structural inference and ligand-protein binding prediction.
  • Pivoted from an underperforming PyTorch GNN to a replacement atom-level Transformer architecture; designed, trained, and evaluated it in one day, producing same-day Pearson correlation of 0.407.
  • Built MLflow governance with preregistered estimates, immutable metric gates, ablations, provenance, and preservation of failed experiments; scientist-facing LLM workflows use XState guardrails, Python Hypothesis validation, deterministic structural-biology checks, and RCSB semantic search.
2025

NeuroNote — Governed Agent Runtime

  • Designed a dual-kernel AI governance architecture that separates generative reasoning from controlled runtime state and deterministic execution, limiting agent authority while preserving useful interaction.
Sodexo · 2025–present

Kitchen Kontrol — Production Operations

  • Built an operational platform with voice capture, structured audit logging, HACCP evidence, and workflow automation designed for gloved, bilingual, mid-service use. Product-vision overlays not in the public GitHub family are not claimed on the case study.
05 · Experience
Oct 2025 — present

Founder & Technical Lead

Eidetix Bio Research

Architected Tokyo Eye, a governed scientific-AI platform for computational drug discovery using hyperbolic GNN and atom-level Transformer/Mixture-of-Experts architectures for ligand-protein binding prediction. Own architecture, data pipeline, model training, and deployment with MLflow governance.

Aug 2025 — present

Area Supervisor & Operations Applications Developer

Sodexo

Translates frontline operational problems into deployable AI-enabled products, combining product ownership, application development, integration design, and adoption.

Jan 2024 — Jan 2025

Technical Business Analyst & Project Manager

International Real Estate Services

Led CRM, analytics, commercial-loan origination, API integration, ETL, process mapping, roadmap planning, and acceptance criteria for audit-ready enterprise workflows.

1999 — 2024

Senior Enterprise Program / Product / Architecture Leadership

Theragun · Sunrun · Mazda · Fox Sports · agencies

Led application modernization, CRM, e-commerce, data, integration, and platform programs for Slumberland Furniture, Sunrun, Mazda North American Operations, Fox Sports, Theragun, Lakeshore Learning, Kayne Anderson Rudnick, and HomePlus Mortgage. Directed cross-functional teams, platform/vendor evaluation, source-of-truth design, governance, rollout planning, and production adoption in complex Fortune 500 environments.

06 · Education & credentials

M.A., Humanities

California State University, Dominguez Hills

B.A., Economics

University of California, Irvine

Certified ScrumMaster (CSM)

Scrum Alliance

Download PDF
01 / 06 · Summary