AI Engineering & Validation

I build AI systems people can actually trust.

Full-stack engineer who orchestrates AI agents to ship faster without sabotaging code quality — with a background in HL7 healthcare interoperability that most AI-focused engineers don't have.

What I do

Three disciplines, one practice.

Not a laundry list of technologies — a specific stack of skills that reinforce each other.

AI Orchestration & Validation

Coordinating multi-agent systems (Hermes, OpenClaw) to build across a full stack without letting output degrade into unreviewed "vibe coding" — plus hands-on work running local LLMs (AnythingLLM, Ollama, LM Studio) to understand serving, parameters, and hardware tradeoffs first-hand.

Healthcare Interoperability

Years of production HL7 interface engineering — building and maintaining clinical data pipelines with Mirth Connect, and more recently moving to native JS/REST integrations to move faster without breaking compliance.

Full-stack & systems

PHP/Laravel, Node.js, React/React Native, Flutter/Dart — monorepo architecture, RBAC, MFA and Supabase Auth, and the infra underneath it, so an AI feature actually survives contact with real users.

Selected work

Case studies

Deep dives, not a project list. Details below are placeholders — drop in your real case studies and these slots update directly.

CASE STUDY 01

Orchestrating AI agents to build a healthcare ecosystem end-to-end

AnythingLLMReact NativeMonorepoSupabase Auth

Problem: a private healthcare product needed an on-prem LLM server, a mobile app, an API, and an admin portal built together — without AI-assisted development collapsing into unreviewed "vibe coding."

Approach: orchestrated Hermes and OpenClaw agents across a single monorepo, enforcing OOP structure and closely auditing every agent output. Layered in MFA and Supabase Auth so speed never came at the cost of security.

Outcome: faster shipping cycles with fewer bugs reaching production than a fully manual build would have taken.

CASE STUDY 02

Modernizing HL7 interface engineering for a clinical data pipeline

HL7Mirth ConnectREST APILIS

Problem: laboratory information system (LIS) integrations were built and maintained on Mirth Connect — reliable, but slow to extend as integration demands grew.

Approach: moved a significant share of new integration work from Mirth to native JS + REST API interfaces, deliberately structured to take advantage of AI-assisted coding speed without sacrificing HL7 compliance.

Outcome: [swap in — turnaround time per integration, number of interfaces migrated, etc.]

CASE STUDY 03

Running local LLMs to understand what actually happens under the hood

OllamaLM StudioAnythingLLMLocal inference

Problem: using AI APIs alone doesn't teach you why a model behaves the way it does, or what it actually costs to run one.

Approach: stood up local LLM servers across AnythingLLM, Ollama (CLI), and LM Studio for different applications — studying how parameter count, quantization, and hardware constraints trade off against each other in practice.

Outcome: firsthand understanding of LLM serving that now informs how agent workflows and validation are designed, rather than treating models as an opaque API call.

Background

Almost a decade in production systems — who’s counting?

  experience.log
2021—nowLead HL7 Interface Developer, IT company (PH) — clinical integrations, coding standards, team lead
2019—2020Senior System Analyst/Developer, private hospital (PH) — interface scripts, Mirth Connect, data warehouse architecture
2019Systems Developer, IT retail solutions (PH) — PHP POS systems
2017—2019Systems Developer, non-profit (PH) — in-house web apps, led a team of 2

Other shipped work

ACS

Mobile app, Play Store — community organization platform

bAdmin

Open-source admin panel boilerplate — GitHub

Gunita

Mobile app, Play Store

Roadmap (on-going)

IN DEVELOPMENT
Driver's Companion App

A mobile companion for drivers — trip logging, vehicle health, and offline-first sync.

FlutterOffline-firstAI poweredCollaborative
How I validate AI

Trust is a testable property.

This is the part most portfolios skip. Here's what "AI validation" actually looks like in practice.

  validation_report.log
eval-suite/Behavior tested against real failure modes, not just happy-path prompts
guardrails/Output constraints and fallback paths for when the model gets it wrong
regression/Re-run evals on every prompt or model change before it ships
observability/Logging and review loops so failures surface in production, not from user complaints
Stack

What's under the hood

Grouped by what each layer is actually for — not just a badge wall.

AI / Orchestration

HermesOpenClawAnythingLLMOllamaLM Studio

Languages & interop

PHPJS / HTML / CSSXMLHL7Dart

Frameworks & platforms

LaravelNode.jsReact / React NativeFlutterSupabase

Infra & deployment

CloudflareVercelGCPDocker