Where I've worked

I joined SIAEC's HR Technology team as an intern. The access governance system technically worked - but only because it was being held together by individual exceptions patched onto individual users. No coherent architecture underneath. An internal audit was coming, and there was no playbook for closing the kind of findings it was going to raise.

I built the architecture the system should have had. New roles designed from the ground up, governance workflows automated, the exception model replaced with something that could actually scale. When the audit arrived, we had the infrastructure to close its findings rather than just acknowledge them. The team offered me a full-time Executive role before the internship ended.

The part I'm most proud of isn't the architecture. It's that I also caught a remediation that had quietly stalled - a batch of access removals submitted weeks earlier that had never been actioned. Nobody noticed. I noticed. That's the work.

Medibles had been running a health assessment on their website for years. Thousands of users had answered questions about pain, sleep, anxiety, and what medications they were taking - all in exchange for personalized product guidance. The data existed. There was simply no way to read it.

I built the entire analytics layer from scratch: the database views that transformed raw WordPress form submissions into something queryable, the data models that made sense of nested JSON fields, the dashboards that turned patterns into decisions. The most useful thing I found: more than half of users were dropping off at the very first personalized question. The product had a funnel problem, and now there was evidence to prove it.

They extended the internship for an additional year to keep building. I take that as the review that matters.

Six months of electronic medical records from one of Los Angeles's busiest emergency departments. Timestamps for every step of the patient journey - registration, triage, waiting, treatment, discharge - but messy, non-linear, and not remotely designed for process analysis. The department knew its length-of-stay was too long. It didn't know where the time was going.

I built a mathematical model of the entire patient flow. Arrival rates by hour of day. Service time distributions fitted to each stage. Routing probabilities at every branch in the process. For one stage - triage - the service time couldn't be measured directly, so I used queuing theory to recover it from wait time data alone. It's one of my favorite things I've ever done analytically.

The answer was counterintuitive. The bottleneck wasn't at the front of the process, where everyone assumed it was. It was further in. The right question turned out to be about what happened after triage, not before it.

KiteSense made an adaptive learning platform. The engine was genuinely good. But it had been built for general academic subjects, and they wanted to expand into aviation - an industry where "learning" is entangled with regulation, logbooks, flight debrief data, and the kind of compliance requirements that make a generic LMS look naive on contact.

I designed the aviation vertical: information architecture, instructor and cadet personas, the integration points with pilot records systems and logbooks aligned to FAA regulations. The part that required the most judgment wasn't the design - it was the B2B pilot with Embry-Riddle. The first proposal was ambitious and expensive. I deliberately de-scoped it into something ERAU would actually say yes to: a focused three-module English prep program. Then I wrote the entire curriculum, authored the question bank, and delivered it as a formal course outline.

The instinct to narrow a pitch into something purchasable is, I've learned, harder than it sounds.

ERAU Asia had six years of admissions data and no clear picture of what it was telling them. Why were local applicants declining? Which recruitment events were actually converting? What did students want from a new campus? Three questions, no answers, and a scattered collection of Excel files that nobody had ever properly merged.

I consolidated the data using R, built an interactive dashboard the enrollment team could actually operate without me, analyzed the admissions funnel across two tracks and identified where the gap was coming from, and benchmarked the institution against ten competitors. The last deliverable was a campus planning survey I wrote from scratch, fielded myself, and analyzed in full.

A second contract followed the first. That's how I knew the work landed.

Azul Airlines was shipping aircraft parts from a central warehouse in Campinas to maintenance stations across Brazil every day. The process worked fine for scheduled maintenance - the parts were known in advance, the logistics were manageable. For unplanned repairs, the kind that happen on the day and need parts now, roughly 40% of shipments were going unused. The material would travel to the station, sit there, and get shipped back.

The problem was that non-routine maintenance was being handled reactively: someone identified a need on the same day a task had to be completed, which gave logistics only hours to respond. There was no model for where non-routine demand was actually going to occur, so there was no way to pre-position parts intelligently.

I built that model - a Distribution Requirements Planning approach using historical consumption data, flight network patterns, and safety stock methodology at the station level. The goal was to shift from same-day scrambling to informed pre-positioning. The work didn't fix the problem by itself, but it gave the Material Planning team the analytical foundation to start.

In 2020, during a pandemic, I started an organization to help Burmese students navigate overseas university applications. I called it COMPASS Learning Myanmar. By 2022, I had incorporated it as Pathway Plus Private Company Limited by Guarantee - registered with Myanmar's Directorate of Investment and Company Administration, with a company constitution, a tax identification number, and a proper governance structure.

I built eight departments, recruited a team of consultants, launched multiple service lines, and secured partnerships with organizations like Startup Weekend Yangon. I also wrote every governance document: the volunteer handbook, the job description book, the NDA, the service agreements, the privacy policy. Not because I had to, but because I understood that an organization that only works when the founder is present isn't really an organization.

I wound it down in 2024. The hardest part of building something is deciding when it's done.