Swan Htet Aung · Technical Product Leader in AI

> story

I started building products in Myanmar back in 2013—back when most people there were just getting their first smartphones. Built one of the country's first anonymous social networks, an app generator for small businesses, mobile experiences that nobody had seen before.

Then came the hard problem: how do you build AI for languages that barely exist online? Burmese, Thai, Khmer—languages with millions of speakers but almost no training data. So we built it from scratch. Created the first NLP technology for Myanmar. Trained models. Built datasets. Made it work.

That turned into EXPA.AI—a conversational AI platform that ended up serving Samsung, Unilever, Nestlé. Millions of conversations. Real customer service automation. The kind where you're actually nervous every time you deploy because real people depend on it working.

Later built Magicsnap—generative AI for brand campaigns. The "turn your selfie into a character" thing, except making it work at scale, for banks and telcos running massive campaigns. Over a million people used it. Had to learn everything about GPU optimization the hard way during our first big spike.

Now I'm in Austin. Spent nearly a decade building and scaling AI products in Southeast Asia. Ready to bring that experience to teams solving bigger, harder problems. Not looking to start another company—looking to join one where I can focus on just building exceptional products.

The Unexpected Turn

Here's a story about plans changing.

I was sitting in Dubai airport last year, waiting for a connecting flight to the US for what was supposed to be a quick business trip. Then I got the news—Myanmar's military announced forced conscription. Going back wasn't an option anymore.

What do you do when "home" suddenly isn't safe? When a business trip becomes permanent? I applied for Temporary Protected Status in the US. Started rebuilding from scratch—again.

But here's the thing about building products in challenging environments: you develop skills that become advantages elsewhere. When you've built AI for low-resource languages with zero datasets, solving problems with proper infrastructure almost feels easy. When you've scaled systems with unreliable internet and power, you build resilience into everything.

A few months ago, I got my EB-1A visa approved—"extraordinary ability" in AI and entrepreneurship. From TPS to permanent residency. It's surreal to have the US government officially recognize the work we did building tech in Myanmar.

Constraints don't limit you. They force you to get creative. And that creativity travels with you wherever you go.

> how_i_work

Start with the problem, not the tech

I've shipped products that failed spectacularly because we fell in love with the technology. Now I start with: what's the actual user pain? Will this solution actually get used? Can we measure if it's working?

Ship small, learn fast

The best product decisions come from real usage data, not conference rooms. Build the smallest thing that tests your hypothesis. Watch what happens. Iterate or kill it.

Own the outcomes, not the output

Shipping features is easy. Moving metrics is hard. I care about whether containment rate went up, whether support tickets went down, whether users actually completed the flow. That's what matters.

Build for maintainability

Technical debt isn't just code—it's poorly documented decisions, missing runbooks, unclear ownership. I've been on-call at 3am enough times to care deeply about operational excellence.

> things_i've_built

A few projects worth mentioning:

Conversational AI at Scale

Built and scaled EXPA.AI to handle 100M+ conversations for Fortune 500 companies. The interesting challenge wasn't the AI—it was making it reliable enough for Samsung's customer service, fast enough for FMCG use cases, and accurate enough for insurance claims.

Real problem solved: How do you build conversational AI that doesn't embarrass your clients in production?

Generative AI for Brands

Magicsnap let brands run "AI selfie" campaigns—upload your photo, get turned into their character. Sounds simple. Making it work for a telco's million-user campaign without melting your infrastructure? That's the interesting part.

Real problem solved: How do you fine-tune diffusion models for brand consistency while keeping render times under 30 seconds?

First Myanmar NLP Technology

When we started, there was basically no Burmese language AI. No datasets, no models, no nothing. So we built it—annotation pipelines, error taxonomies, the whole research-to-production loop. Led a 40+ person engineering team to make it happen.

Real problem solved: How do you build NLP for low-resource languages from absolute zero?

> what_i_believe

Most AI products fail because of product problems, not AI problems. The model works fine—but nobody wants to use it, or it solves the wrong problem, or the UX is terrible, or it's too slow for the actual workflow.

The best product people are bilingual. They speak "technical" well enough to have real architecture discussions with engineers, and "human" well enough to understand why users are frustrated. That translation layer is where most products break down.

Enterprise AI is about trust more than accuracy. You can have a 95% accurate model, but if you can't explain why it made a decision, or if it occasionally says something wildly wrong, enterprises won't deploy it. Reliability and auditability matter more than being cutting-edge.

Good product work is mostly about saying no. Every team has infinite ideas. The hard part is figuring out what not to build, what to kill, what to simplify. Shipping less, but better, is underrated.

> what_i'm_doing_now

Currently in Austin, exploring opportunities with companies doing interesting work in AI products—particularly around conversational AI, LLM applications, or anything in the healthcare/enterprise space.

Interested in roles where I can bring my technical product background to complex problem spaces. Not looking to start another company—looking to join teams where I can focus on building exceptional products without being in every board meeting.

If you're working on hard problems in AI products and think my background might be helpful, let's talk.

> get_in_touch

→ Email me → LinkedIn → Call me