Things I've made

Not a list of achievements. Just projects with stories — challenges faced, things learned, and (hopefully) lessons drawn.

I prefer telling process over outcome. Each project teaches something — sometimes technical, often about myself.

🛠Project

2021-Now

DeFiDataAIResearchLeadershipProduct

Kyber Network — From Research to Data Ownership

The Challenge

Democratize DeFi trading intelligence through AI, then build and own the entire data function for a DeFi protocol from scratch — while the protocol was live, scaling, and launching new products.

The Process

Started as Researcher (2021), became Research Lead (2022). Led KyberAI (2023) with the team — an ML product for trading insights across 4000+ tokens. Colleagues jokingly called me "Father of KyberAI." Product sunset in Dec 2023, but the lessons stayed. Became Data Owner (2025): built data infrastructure from scratch with a small team, migrated from unreliable third-party sources to in-house pipelines, supported major product launches. Daily work spans data product ownership, analytics, and research support.

The Lesson

Products can be discontinued but the learning remains. Building something from 0 to millions of users teaches you things no course can. Building reliable data infrastructure is 80% unglamorous work — but when teams stop asking "can we trust this data?" and start asking "what can we do with it?" that's when you know you've succeeded.

The Impact

KyberAI taught me that discontinued products aren't failures — they're education. The data infrastructure work taught me patience: no one notices when things work, everyone notices when they break. Still learning what it means to own something end-to-end.

Learn more→
○
🛠Project

2017-Now

AINLPStartupVietnam

EM&AI — Vietnamese AI Startup

The Challenge

Build AI tools for Vietnamese language — a low-resource language with unique characteristics that most global AI tools handle poorly.

The Process

Started while doing PhD (2017). Early days: debates with engineers who insisted their algorithms beat AI for NER — they were right, then. Built Vietnamese NLP toolbox (SVM, CNN), ASR systems, speech synthesis. Evolved from CTO to Scientific Advisor.

The Lesson

Building a company while doing a PhD is not recommended. But sometimes the best learning comes from doing too much at once. Also: never bet against AI progress.

The Impact

Still running after 7+ years, which is the real victory for any startup. Watching it grow from a distance taught me that sometimes the best contribution is knowing when to step back.

○
📺Media Feature

2014-Now

AIJourneyLearningPersonal

A Decade with AI

The Challenge

Stay curious through multiple AI winters and summers. From reading a whole book on Lagrange formulas to understand SVM (2014) to vibe coding with Claude (2026).

The Process

2014: Student reading math to understand ML. 2017: NLP debates at Vietnam AI startup. 2017: Co-founded EM&AI. 2021: Joined Kyber Network as Researcher. 2022: Research Lead. 2023: Led KyberAI. 2023: GPT Sputnik moment. 2025: Data Owner. 2026: Building personal projects, motivating colleagues to use AI, vibe coding this website.

The Lesson

Each era feels like the pinnacle. Each is just the beginning. The best time to start with AI was 10 years ago. The second best time is now.

The Impact

Still learning. Still building. Still curious. Now motivating everyone around me to embrace AI tools.

○
📄Research Paper

2020

PhDSecurityMLBayesian Networks

PhD: Security for Named Data Networking

The Challenge

How do you detect attacks on a network architecture that doesn't exist yet? NDN was (and is) experimental — no real deployment data to learn from.

The Process

Built testbeds, simulated attacks, developed anomaly detection with Bayesian Networks. Discovered something backwards: network effects can improve ML. Broke down Bayesian Network functions into lambda calculus — an idea waiting for IoT/distributed AI to mature.

The Lesson

Academic research moves slowly. Some ideas need to wait for their time. The lambda calculus work may matter more in 10 years than it did in 2020.

The Impact

The publications are nice. But the real value was learning how to think about problems that don't have solutions yet. Also learned that peer review, while painful, makes everything better.

Learn more→
○
📄Research Paper

2017-2020

EUResearchSecurityNLP

European Research Projects

The Challenge

Contribute to large-scale EU research initiatives while maintaining focus on PhD thesis. Multiple projects, multiple deadlines, multiple countries.

The Process

DOCTOR (anomaly detection with BNC), INSPIRE-5Gplus (NLP for security policies), DigitBrain (defect detection with ML), Mosaico (security orchestrator). International collaboration across time zones.

The Lesson

EU research bureaucracy is real. But the network of researchers you build is worth every deliverable report.

The Impact

The research deliverables are archived somewhere. The friendships and collaborations across Europe — those are still active. Turns out that's what matters.

This collection is incomplete — and always will be. Each project teaches something new, adding another artifact to this growing museum.