Systems Architect → AI Platform / MLOps Engineer
Lewisville, TX · bdebnam@gmail.com · linkedin.com/in/bryanhd
Resume — AI Platform / MLOps:
PDF ·
DOCX
| Systems Architect:
PDF ·
DOCX
Positioning
20+ years in enterprise IT, 3+ at the architecture level — cloud migrations, M&A technology
integration, and identity modernization across four acquisitions, plus building and leading a
60+-control SOX/ITGC compliance program.
At work, more recently: “Script Portal,” an internal platform for interchangeable
agentic-AI backends under one allowlist + output-masking safety model; Playwright RPA agents
for SOX audit evidence; the org’s GitHub Enterprise Copilot rollout.
Independently (personal R&D; foundation: Texas McCombs AI/ML post-grad, 2024–2025): a full
MLOps governance lifecycle (model registry, champion/challenger promotion with rollback,
automated drift detection, data lineage), an LLM security triage tool, and LLM cost analytics —
all on a self-hosted, dual-GPU Incus/k3s fleet.
What’s distinctive: the intersection of SOX/ITGC audit discipline and hands-on AI-platform
engineering — few people have both.
Selected projects
Each project is tagged [Work] (Inform Diagnostics) or [Independent R&D] (personal), and
scoped: Production (in real organizational use), Reference build (complete and running,
single-environment / not scale-tested), POC (proves the pattern).
An end-to-end model-governance pipeline for a 7-agent LLM system, applying SOX/ITGC evidentiary
rigor to AI oversight.
Model registry & staged promotion (MLflow) — preflight validation gates, automated
rollback, and a silent-no-op safety guard that blocks unvetted model changes from reaching
production.
Champion/challenger A/B gate — a staged candidate model is evaluated head-to-head against
the live champion before any cutover.
Automated drift detection — weekly scheduled scanning, parameter-order-insensitive.
Caught a live defect in production: an arithmetic self-contradiction in a prompt-scoring agent
that was inflating its own output score.
Data lineage (Marquez / OpenLineage) across the MLflow ↔ Ollama pipeline — every model
version, eval run, and promotion decision is auditable.
LLM-as-judge evaluation harness — scores agents on substance (accuracy, completeness, no
fabrication) rather than surface grammar; registered judge model; testsets logged as MLflow
GenAI datasets. Brought all 7 agents to a validated quality gate.
AI Gateway (FastAPI/uvicorn) — unified model-access boundary with a local detect-quality
model; GenAI UI wired (Prompts registry for 7 agent system prompts, review queues, label
schemas).
2. Provider-Agnostic Agentic AI Platform (“Script Portal” / Agent Bridge) — [Work] · Production
An internal platform for deploying interchangeable agentic-AI backends under one safety model.
Interchangeable agent CLIs — Claude Code CLI, Copilot GPT CLI, and a custom VS Code
“Agent Bridge” extension (LLM-broker) — behind a shared allowlist and output-masking layer.
Secret-aware execution framework; provider-agnostic orchestration so the underlying model/
vendor can be swapped without changing the agent contract.
Playwright-driven RPA agents for SOX evidence gathering — agents authenticate to
enterprise portals, make record-selection decisions, and capture timestamped, masked evidence
into the correct control package.
Built and led the SOX/ITGC compliance program — 60+ control activities, PowerShell-
automated evidence generation; primary liaison between Infrastructure, Security, and Audit.
Manage the org’s GitHub Enterprise Copilot deployment; built an n8n workflow monitoring
seat utilization — reclaimed 25 inactive seats (~$475–725/month).
Led M&A technology integration for four acquisitions — M365 tenant consolidation, AD/Entra
ID identity integration for ~1,800 users, 400+ mailbox/OneDrive migrations, multi-forest AD
trust across 6 forests.
Skills — evidenced vs. currently deepening
Area
Evidenced (projects above; W = work, R = independent R&D)
Microsoft(active through 2027): Azure Solutions Architect Expert · Azure
Administrator Associate · Azure Fundamentals (2019) · AI-102 Azure AI Engineer
Associate — in progress
Post Graduate Program in AI & Machine Learning: Business Applications — McCombs School
of Business, The University of Texas at Austin (conferred July 2025)
A.A.S., Information Technology — CIT: Network Administration & Support — Dallas County
Community College District (now Dallas College)
U.S. patent (2014 Microsoft hackathon project, placed 25th worldwide)
Honest scope statement
Two separate evidence bases, never blurred:
Independent R&D (the sandbox) is a single-host reference platform, not a production
system with a user base — it demonstrates architecture judgment, integration breadth, and
knowledge of failure modes. Projects 1, 3, 4, 5.
“At scale / in production” evidence comes from the work at Inform Diagnostics: four
M&A integrations, identity modernization for ~1,800 users, a 60+-control SOX program, an
org-wide Copilot rollout, and the Script Portal agent platform in real internal use.
Projects 2, 6.
Verification
Issuer-hosted records, verifiable without an account:
Microsoft Certified: Azure Solutions Architect Expert — verify on Microsoft
Learn
· earned Nov 2019, active through Nov 2027
Microsoft Certified: Azure Administrator Associate — verify on Microsoft
Learn
· earned Apr 2019, active through Oct 2027