Phat Nguyen Thuan GitHub

AI · tools · research · side quests

I’m learning
by building.

Hey, I’m Phat. I study AI at UIT, work as a Junior AI Specialist at Amaris, and build little tools whenever something annoys me often enough. This is where I keep the projects, research, talks, and experiments I’m learning from.

See selected work
HCMC, Vietnam Usually curious about how things work

Curious by default.
Still figuring things out.

I like building things that make a complicated idea feel a little easier to use.

Most of my projects start with a question, a small annoyance, or one tab too many. I work on agents, memory, model evaluation, GPU inference, and learning tools—but the common thread is simple: I want the result to be useful to a real person.

I’m studying AI at UIT in HCMC and working as a Junior AI Specialist at Amaris. Before that, I worked on ByteRover’s agent-native memory tools and open source. I’m still learning a lot, and I like sharing the useful parts as I go.

2023—27
Studying AI University of Information Technology
07/26—now
Working as a Junior AI Specialist Amaris Consulting
07/25—06/26
Before that: AI Engineer & Developer Advocate ByteRover
Figure it out Make it work Share what I learned

I like to know
how things work.

The projects change, but I keep coming back to the same questions: can it be useful, can I understand it, and can I explain it to someone else?

How do we get models
to do useful work?

Agents, MCP, memory, and evals—the practical bits that help models use tools without turning the whole system into a mystery.

What’s happening
under the hood?

Inference speed, KV-cache behavior, batching, quantization, and GPUs—the stuff that decides whether an idea works outside a notebook.

How do we make an idea
click?

Multimodal research, learning tools, explainers, and small playful products that give a technical idea somewhere to land.

Stuff I’ve been
working on.

Here’s a small slice of my GitHub. The projects move between agent systems, research, learning tools, and whatever else I’m curious enough to try.

There’s more in the archive: course helpers, research forks, macOS utilities, and a few things built just because I was curious.

Browse all repositories

I like asking
what actually works.

A few places where curiosity turned into a paper, a benchmark, or a very long debugging session.

01
2026 · co-author

ByteRover: Agent-Native Memory Through LLM-Curated Hierarchical Context

A paper about giving agents memory that understands the context it is saving, rather than treating memory as a separate black box.

02
2025 · co-author

ROOMELSA — language and spatial guidance for 3D object retrieval

We explored how language and spatial clues can help retrieve the right 3D object. The team took first place in the track.

↗
03
2025 · co-author

A hybrid video retrieval system using CLIP and BEiT-3

A hybrid visual-and-text search system for finding events in video. It earned second place at HCMC AI Challenge 2024.

04
2025 · computer vision

Robust traffic vehicle detection

An object-detection project focused on real-world mistakes, especially vehicles and pedestrians getting mixed up. Second place at SoICT Hackathon 2024.

I learn it.
Then I try to share it.

2025
OpenInfra & Cloud Native Day Vietnam

I talked about benchmarking LLM performance across GH200, H100, and A100 systems.

↗
2025
HCMC AI Meetup #1 · Menlo Research

A practical session on training with NVIDIA GH200 and getting comfortable with ARM64.

2026
Third Prize · Codex Hackathon HCMC

My team and I built LitMatch, a more playful way to meet Vietnamese literature.

Let’s make something
useful—or weird.

If you’re curious about agents, education tools, AI systems, or a side project that refuses to behave, feel free to say hi. I’m always happy to compare notes.

LinkedIn GitHub npm