quoc@portfolio:~$

> whoami

Data systems · Applied AI · Engineering tools

Ton Dai Quoc

Data / AI Engineer

I build data systems that move from ingestion and transformation to models, APIs, observability, and interfaces people can actually use.

inspect work view résumé
location
Ho Chi Minh City, Vietnam
Ton Dai Quoc
focus://data-ai-engineering--.- ..- --- -.-.

> cat positioning.txt

Engineer first. Model second.

I care about the full path around a model: trustworthy data, reproducible workflows, observable services, and a delivery surface people can use.

availability

Actively looking for a Data or AI Engineer position, internship, or research collaboration in Ho Chi Minh City or remote. Available for technical interviews and take-home work on request.

start a conversation

01Selected systems

Work that runs end to end.

The strongest builds are shown as compact case studies. Every stack entry below is also used to generate the skills matrix.

Capybara project cover placeholder
Developer Tool2026Solo engineerOpen source

A terminal-native trace debugger for AI agents.

Capybara

A Go CLI that records agent runs from OTLP endpoints, coding-agent sessions, and JSONL files, then finds tool failures, output drift, loops, improvised answers, and cost spikes.

failure detectors
7
trace ingestion
OTLP
debug workflow
TUI

system notes

  • Cost-sorted waterfall, run diffing, and turn-by-turn inspection.
  • Replay with edited tool output to test alternate execution paths.
  • Local-first storage with SDK and OpenTelemetry integrations.

runtime / stack

  • Go
  • SQLite
  • OpenTelemetry
  • OTLP
  • Python SDK
  • Node.js SDK
  • TypeScript
open repository

Shopee review intelligence

One system across three repositories: the crawler collects products and reviews, the PhoBERT model scores them by aspect, and the hub serves both to an operator dashboard.

Shopee Scraper project cover
Data Platform2025Solo engineer

Shopee Scraper

Resumable collection for product and review datasets.

crawl modes
3
balanced sampling
5*
browser control
CDP
  • Python
  • nodriver
  • Typer
  • Pydantic
repository
ABSA PhoBERT project cover
AI / ML2025Solo engineer

ABSA PhoBERT

Vietnamese aspect-level sentiment for product reviews.

labeled reviews
8,424
micro-F1
0.91
aspect / sentiment
8x4
  • Python
  • PyTorch
  • Transformers
  • PhoBERT
  • scikit-learn
  • Pandas
repository
E-Metric Hub project cover
Product2025Contributor

E-Metric Hub

Shopee operations and review intelligence in one product.

review analysis
ABSA
commerce dashboard
KPI
stack delivery
FULL
  • Python
  • PhoBERT
  • React
  • Redux
  • Node.js
  • REST API
repository
KnightVision project cover placeholder
Data Platform2026Solo engineerOpen source

Chess analytics from raw PGN to interactive analysis.

KnightVision

An end-to-end platform that ingests Lichess data, processes Bronze/Silver/Gold layers with PySpark, models analytics with DuckDB and dbt, orchestrates Airflow, and serves a FastAPI + React analysis surface.

medallion layers
B/S/G
source format
PGN
system ownership
E2E

system notes

  • Idempotent ingestion and replay-safe PySpark transformations.
  • DuckDB + dbt analytical layer with tested marts.
  • Optional Kafka path and a custom dashboard for game analysis.

runtime / stack

  • Python
  • PySpark
  • Apache Kafka
  • DuckDB
  • dbt
  • Apache Airflow
  • FastAPI
  • React
  • Docker
  • Parquet
  • Stockfish
open repository
RetailRocket Pipeline project cover placeholder
Data Platform2026Solo engineerOpen source

E-commerce intelligence over 2.7 million behavioral events.

RetailRocket Pipeline

A production-shaped batch data system that reconstructs sessions with point-in-time joins, publishes dbt-tested marts, and serves recommendation plus cart-abandonment models through an API and dashboard.

source events
2.7M
data layers
3
ML workloads
2

system notes

  • Sessionization and point-in-time feature engineering in Spark.
  • ALS + item2vec recommendations and cart-abandonment scoring.
  • Data quality, drift, service, and infrastructure monitoring.

runtime / stack

  • Python
  • PySpark
  • Apache Airflow
  • dbt
  • PostgreSQL
  • FastAPI
  • React
  • Spark MLlib
  • XGBoost
  • Prometheus
  • Grafana
  • Docker
open repository

Applied research

University research that shares one habit: take a published claim, rebuild it end to end, and report what survives. Code for these lives locally rather than in a public repository.

PGE Return Forecasting project cover
AI / ML2026Researcher

PGE Return Forecasting

Training a forecaster on trading utility, not on forecast error.

hero Sharpe
1.09
Optuna trials
200
ablation runs
80
  • Python
  • XGBoost
  • Optuna
  • SHAP
  • Pandas
  • NumPy
  • scikit-learn

repository not public

SGD Sampling Study project cover
AI / ML2026Researcher

SGD Sampling Study

Testing whether smart sampling really beats random sampling.

datasets
4
largest row count
845K
seeds per test
N=10
  • Python
  • NumPy
  • SciPy
  • scikit-learn
  • UMAP
  • PySpark

repository not public

Offside Decision Module project cover
AI / ML2026Module owner

Offside Decision Module

Turning pitch geometry and player pose into an offside call.

pipeline modules
3
pitch coordinate frame
105x68m
pose keypoints
COCO-17
  • Python
  • NumPy
  • OpenCV
  • PyTorch
  • Ultralytics YOLO
  • SoccerNet

repository not public

more work

Football Management System project cover
Product2025Contributor

Football Management System

Desktop operations for running a football league.

  • Java
  • JavaFX
  • FXML
  • Oracle Database
  • Apache POI
  • iText
repository
LFM2 Serving on vLLM project cover
AI / ML2026Solo engineer

LFM2 Serving on vLLM

Cutting time-to-first-token on a shared GPU slice.

  • Python
  • vLLM
  • Hugging Face
  • Docker
repository
Weather Analysis Dashboard project cover
Product2025Contributor

Weather Analysis Dashboard

Forecasts, anomaly detection, and recommendations in one view.

  • JavaScript
  • React
  • Laravel
  • PHP
  • MongoDB
  • Chart.js
  • D3.js
  • Docker
repository

02Capability map

Project-backed stack.

This matrix is generated from the project data source, then augmented with technologies confirmed from the wider profile.

Languages

07 entries
  • Go1 project
  • Java1 project
  • JavaScript1 project
  • PHP1 project
  • Python9 projects
  • SQLprofile
  • TypeScript1 project

Data systems

10 entries
  • DuckDB1 project
  • PostgreSQL1 project
  • MongoDB1 project
  • SQL Serverprofile
  • Oracle Database1 project
  • SQLite1 project
  • Apache Airflow2 projects
  • Apache Kafka1 project
  • dbt2 projects
  • PySpark3 projects

AI / ML

14 entries
  • NumPy3 projects
  • OpenCV1 project
  • Optuna1 project
  • Pandas2 projects
  • PyTorch2 projects
  • scikit-learn3 projects
  • SciPy1 project
  • Spark MLlib1 project
  • Stockfish1 project
  • Transformers1 project
  • Ultralytics YOLO1 project
  • UMAP1 project
  • vLLM1 project
  • XGBoost2 projects

Product engineering

11 entries
  • Chart.js1 project
  • D3.js1 project
  • FastAPI2 projects
  • Laravel1 project
  • Node.js1 project
  • Node.js SDK1 project
  • Pydantic1 project
  • Python SDK1 project
  • React4 projects
  • Redux1 project
  • REST API1 project

Operations

12 entries
  • Docker4 projects
  • Docker Composeprofile
  • Gitprofile
  • GitHub Actionsprofile
  • GoReleaserprofile
  • Grafana1 project
  • Linuxprofile
  • Makeprofile
  • OpenTelemetry1 project
  • OTLP1 project
  • Prometheus1 project
  • uvprofile

03Operating context

How the work gets built.

Trustworthy data, reproducible workflows, observable services, and documentation that stays next to the code.

> cat approach.txt

I'm Quoc, an Information Systems undergraduate in Ho Chi Minh City building data platforms and applied machine learning. I like systems that can be inspected and rerun: explicit contracts, tested transformations, observable jobs, and documentation that sits next to the code.

04Contact

Let's build something measurable.

Open to Data and AI Engineer roles, internships, and infrastructure-minded product work. Send a message or inspect the repositories directly.

github linkedin