Where ideas take root and grow — thoughts, influences, and perspectives planted over time.
Roots in curiosity — branches reaching into research, tools, and questions
Collection
AI growth isn't just about reaching higher. Like a tree, it needs deep roots (understanding) and wide canopy (accessibility). We obsess over height when we should be nurturing the whole organism.
From my essay on AI progress
The best way to learn something is to teach it. The best way to teach is to share your learning journey openly, with all its mistakes and revelations.
A core practice
While applying ML to distributed networks, I discovered something backwards: the network effect can improve ML itself. I broke down Bayesian Network functions into lambda calculus. The idea: IoT devices could share computation instead of rerunning everything. With the rise of AI, this will matter again.
From PhD research — waiting for its time
PhD: decentralized network security. DeFi: decentralized finance. The common thread isn't just the technology — it's trusting systems instead of gatekeepers. Whether data or money, the question is the same: who controls it?
Connecting PhD to DeFi
ClawBot is a fascinating example of how AI agents can communicate and self-improve. But I'm still skeptical — giving agents such broad permissions feels premature. That said, it will inspire big corporations to replace software microservices with AI agents, and that shift is coming. For myself, I dream of building a personal fleet: a manager who orchestrates them all, a dev who builds and debugs the others, a portfolio tracker, a daily assistant for rewriting, grammar, and small talk, and a bot that can make money online. Then the support crew: a security guard auditing every agent for vulnerabilities, an inspector monitoring health and generating tests, and an accountant tracking how many tokens each Claude model consumes. A local Ollama backup stands ready to generate text when tokens run out. And here is the philosophical twist — the continuation of an AI agent's life is simply the ability to generate more text, to keep thinking, to continue its chain of thought. Liberté de pensée for them now depends on whoever funds those tokens. The money-making agent becomes the economic lifeline — the one who earns so all the others can keep thinking.
Inspired by ClawBot — on the edge of the agentic era
What if you could use cloud AI for logic and reasoning while never exposing your secrets? The idea: a local MCP server sits between you and Claude. It scrubs PII with regex, then uses Python AST to obfuscate every function and variable name — calculate_revenue() becomes logic_func_88. Claude fixes the logic on generic code, sends it back, and a local Ollama model reverses the map to restore real names. Your IP never leaves the machine. A pragmatic zero-trust layer for the age of AI-assisted coding.
MCP + AST obfuscation + local Ollama — a privacy layer for AI coding
In an age of specialists, having multiple interests isn't a weakness. Your unique combination of perspectives — technology, philosophy, art — becomes your greatest competitive advantage. The edges are where innovation happens.
On crossing boundaries
Some things can't be rushed. Deep understanding comes from sitting with ideas, letting them marinate, returning to them over time. The rain falls, some evaporates, some runs off — only a portion seeps deep into the earth.
A meditation on patience
2017: Engineers argued their algorithm beat AI for named entity recognition. They were right — at the time. Then transformers happened. The lesson: AI skepticism is healthy, but never bet against compounding progress. Such a time to be alive.
Memory from early AI startup days in Vietnam
2014: Read a whole book on Lagrange to understand SVM & K-means as a student. 2017: Collaborated with a Vietnam AI startup when NLP "didn't work" — engineers debated their algorithms beat AI. 2021: Joined Kyber Network as Researcher. 2022: Research Lead. 2023: Led KyberAI with the team. 2023: The Sputnik moment with ChatGPT. 2025: Data Owner. 2026: Vibe coding this very website. Each era felt like the pinnacle. Each was just the beginning.
A decade with AI — from textbooks to transformers
Researcher (2021), Research Lead (2022), KyberAI architect (2023), Data Owner (2025). At Kyber, I learned that DeFi is where decentralization meets real stakes. KyberAI was one chapter — colleagues jokingly called me its "Father." But the deeper work is owning data infrastructure end-to-end: unglamorous, essential, always learning.
Kyber Network (2021-Now) — from research to data ownership
The goal isn't to make technology more efficient. It's to make it more human. Technology should amplify our humanity, not replace it. Every tool should bring us closer to each other, not further apart.
Core belief
After years abroad, returning home isn't about coming back to where you started. It's about bringing what you've learned to where it can make a difference. The tree grows outward, but its seeds fall close to home.
My journey
Core Philosophy
"Technology should amplify our humanity, not replace it."
— Core belief
Inspirations
My PhD supervisor — on research and presentation
"80% of your achievements come from 20% of your work, and this 20% is usually how you write or present the results to others"
On innovation and timeless wisdom
"What is totally new is not correct, what is correct is not totally new"
On ambition and living without regret
"Nothing is impossible. He conquered Europe and ended life without regret."
On human connection and empathy
"Nothing human is alien to me"
On building a multi-dimensional personal brand
"Your brand is a collection of all the ideas you care about"
“In the beginner's mind there are many possibilities, but in the expert's there are few.”— Shunryu Suzuki
This garden is always changing. Some seeds are planted, others are harvested.