Twelve full-stack
serverless domains with retrieval-augmented AI

A publication gate that fails closed

I took over a stalled, two-year-old heritage platform with fifteen prior contributors and rebuilt its production system in coordination with the Smithsonian and Library of Congress. The original loop could publish an unsupported summary when no attempt passed. I replaced it with a gate where two cross-family scorers must independently reach 90/90, a per-claim citation auditor blocks on uncertainty as well as falsity, scorer disagreement routes to human review, and generation is grounded against a schema-validated ground-truth corpus before it begins. I cleaned and reconciled the full transcript corpus line by line in six days, wrote the team's 21-page technical paper, taught the practice in weekly university sessions, and continue to own the live system and public MCP surface.

The archive brings together 140 verified interviews and was presented at King's College London. The work led to a publication with Oxford University Press.

Retrieval settled by evaluation; governance enforced mechanically

I built the workbench from scratch around 500 CODW files that only expensive commercial software could read, writing the format reader and turning a principal investigator's dissertation methodology and a decade of anonymized interviews into a live application. Retrieval runs through a two-backend adapter (pgvector with full-text search against Qdrant) with the winner chosen by a gold-question evaluation harness rather than preference. Every answer traces to a transcript, turn, and timestamp. Row-level security runs under the caller's identity, while a deny-by-default ignore list and pre-commit guard prevent any transcript or secret from entering Git. Precomputed Postgres RPCs and a per-session cache cut the worst-case analysis path to roughly 0.4 seconds.

Referred by the UCLA principal investigator, I serve as AI Engineering Lead and own the site. I am the designated engineering lead for expansion as the team brings in collaborators and pursues grant funding.

Executive strategy with a measured production rebuild

I own the technology stack for Aloha Carbon, setting AI and infrastructure strategy and advising the executive team across technology, revenue, and operations. I run ongoing hands-on AI coaching for executives, mechanical engineers, and civil engineers. The public engineering artifact is a reverse-engineering of a hacked, plugin-heavy WordPress site into a static React/Preact deployment with a strict Content-Security-Policy, HSTS, hardened headers, and no PHP or application database. The rebuilt surface is live at alohacarbon.org.

Follow the money through the instrumentation

A ten-day data audit exposed a click-hijacking system intercepting 99.1% of promised delivery from the vendor consuming 77% of the client's ad budget. A forensic read of the vendor's public 10-K and 10-Q filings against its represented audience figures made the discrepancy provable rather than suspected. I stripped five redundant tag-manager containers, thirteen deprecated scripts, and forty-eight conflicting third-party integrations, raised PageSpeed from failing to the high 90s, rebuilt campaigns, and delivered a mechanism for locking committed currency exposure eight to ten months forward.

Compliance first; automation second

I documented system boundaries, mapped access control, wrote workstation-hardening and audit-readiness procedures, built scorecards, and delivered security training to clinical and administrative teams. Once baseline controls were established, I shipped operational workflows across patient intake, scheduling, content generation, and retention, then expanded operating margins with an AI-driven campaign that supported a lease on an additional location. For protected data that cannot leave the premises, I designed split inference: local nodes handle patient information and only de-identified secondary processing crosses an API boundary.

Research turned into repeatable infrastructure

I ran institutional infrastructure, DevOps, and quantitative research systems for a hedge-fund pre-launch, coordinating prime brokers and custodian banks on logistics including overnight distributed cash sweeps. I triaged a 200-entry parameterized risk playbook into a working set of 180 structures, each tied to a stated market condition, built a multi-GPU tensor-parallel inference pipeline for bounded financial tasks, and established repository controls, CI/CD checks, and evaluation harnesses. When seed funding fell through, the work continued publicly as a substantially stronger AI Gamma platform.

Architecture, mathematical review, and product delivery

During my 2023–2025 tenure as Principal Architect, I designed the modern platform and its models, directed all major product and quantitative analytical tool releases, and modernized the earlier key-level software into a coherent set of models and tools. I served as the primary subject-matter expert on market mechanics, derivatives modeling, and product development for the engineering and product teams, carrying responsibility for the mathematical soundness of what shipped. I also owned product education and technical escalation, produced 585 consecutive market-close newsletters and a 700-page market-mechanics knowledge base, and answered more than 10,000 client questions directly.

Engineering direction across a global virtual estate

As Senior Systems Engineer 5 and Lead Infrastructure Architect, I headed DCMA West's virtual estate across 22 countries. I led hardening, migration, and optimization, moving thousands of active virtual machines from legacy clones to Horizon View Instant Clones on budget-constrained vSAN clusters. I charted the agency's first end-to-end network topology, bringing thousands of virtual machines across 20 global datacenters into an interactive map for department heads. During production rollouts, I provided engineering direction, hands-on troubleshooting, and technical risk assessments for Microsoft and VMware engineering teams.

A formal appointment with direct cybersecurity responsibility

I was formally appointed Information Systems Security Officer for the MARFORPAC G-6 headquarters virtual-server environment, serving as the technical cybersecurity advisor and reporting directly to the Branch Head. I led security-hardening and policy-compliance work alongside the base cybersecurity department and threat red teams, and managed federal auditors through a two-week review that earned a passing compliance score. I also restored a broken VMware upgrade path, wrote the engineering documentation for future enterprise upgrades, and handled storage scaling, PKI certificate overhauls, backups, and incident response during live outages and cyber incidents.