Turning raw data into decisions people act on.

• End-to-End ETL Pipeline

  dbt, Airflow and Docker automating data workflows into BigQuery, with automated data-quality tests.

  Result — replaces manual data pulls with a scheduled pipeline; quality tests run before anything reaches the dashboard.

  Data Engineering ↗
• Daily Business Dashboard

  Interactive Power BI dashboard for daily campaign performance and revenue tracking.

  Result — one dashboard replaces the daily manual pull-and-format routine for campaign and revenue numbers.

  BI & Visualization ↗
• BigQuery Table Automation

  Bulk-creates flattened event tables on BigQuery in minutes, streamlining preprocessing.

  Result — a multi-step manual flattening job becomes one script run, with daily incremental updates after that.

  Data Engineering ↗
• Auto Marketing Report

  Automated weekly marketing report via Gmail API and Selenium, start to finish.

  Result — ships six brand/market reports a week with zero manual steps, and holds the run instead of shipping data with missing SKUs.

  Automation ↗
• IAP User Prediction

  ML model predicting in-app purchase behavior from user login patterns.

  Result — flags likely payers early enough to act on, from login-pattern signals alone.

  Machine Learning ↗
• User Behavior Analysis

  Statistical analysis of IAP user behavior to identify key engagement drivers.

  Result — traced engagement swings back to specific behavioral drivers, not just a headline number.

  Analytics ↗
• Predict Campaign ROAS & LTV

  ML model predicting ROAS and LTV for UA campaigns at D3, D7, D14 and D30.

  Result — gives UA spend calls a forward-looking number at D3, instead of waiting until D30 to know if a campaign worked.

  Machine Learning ↗
• Pay Rate Analyst Check

  Diagnosed the drivers behind pay-rate changes, end to end.

  Result — traced a pay-rate swing back to its actual driver instead of leaving it as an unexplained number on a report.

  Analytics ↗
• Amazon Ads & Sales Performance Analyst

  Five-module analytics suite for an Amazon seller account — keyword scale/kill calls, weekly sales diagnostics, demand and inventory forecasting, and A/B-test analysis, on data built to mirror how Amazon actually reports numbers.

  Result — backtested three forecasting approaches and shipped whichever won on the backtest, not the fanciest one; the ML model lost to Holt-Winters.

  Analytics ↗
• Mobile Game Analyst

  Five-module live-ops analytics suite for a level-based mobile puzzle game — difficulty as the core lever for retention and monetization, from level design up through UA spend decisions.

  Result — onboarding funnel diagnostics hit 96.6% agreement against a hidden ground-truth label, with every miss traced to one specific mechanism.

  Machine Learning ↗
• E-Wallet Churn Prediction: Trees vs DNN vs LSTM/GRU

  Binary 6-month churn model for a Vietnamese e-wallet — benchmarking tree/ensemble baselines (LightGBM, XGBoost, Random Forest) against DNN, LSTM and GRU, plus when the model actually needs retraining.

  Result — RandomForest/XGBoost/LightGBM/DNN all land within ~1pt AUC (0.68-0.69); LSTM/GRU trail well behind (AUC ~0.52) from a documented input-signal gap, not a training bug. Calibrated the decision threshold to hit 92% recall on churn (from 67% at the default 0.5) — the deliberate precision trade-off this retention use case calls for.

  Machine Learning ↗

Core stack — Python · SQL · Pandas · Power BI · BigQuery · dbt · Airflow · Docker · PyTorch

• Fintech & Credit Risk — Viettel Digital Service

  Data platform reliability and credit-risk feature engineering for a lending product — building the infrastructure that catches data problems before they reach a credit-scoring model, not just the model itself.

  Designed a self-hosted observability platform (ClickHouse + Airflow) with statistical drift detection across 150+ feature columns, at zero license cost — it caught a real production incident, including a full week of missing data volume and a significant null-rate spike, before either reached the model.

  Drift detection, illustrative

• Mobile Gaming — FPT AdOne, PixOn Game Studio

  Live-ops analytics for a mobile game studio — game economy and retention tuning backed by Looker Studio dashboards, statistical testing, and ML models predicting which players convert and what they're worth.

  Built classification models predicting which players become payers and regression models forecasting their LTV — turning game-economy tuning into a testable hypothesis instead of a gut call on balance patches.

  Payer prediction, illustrative

• E-commerce & Digital Marketing — BlueStars LLC

  Campaign and inventory analytics across e-commerce marketplaces — Power BI reporting paired with regression and classification models turning ad spend and stock decisions from reactive into anticipated.

  Regression models forecasting sales and inventory needs, paired with classification models flagging which products were likely to succeed — before the ad budget was committed, not after.

  Success split, illustrative

• Mar 2026 — Present
  Data Analyst · Viettel Digital Service

  Designing data platform reliability and credit-risk feature pipelines end to end — a self-hosted observability prototype on ClickHouse + Airflow, and feature engineering for a credit-scoring model.

• Mar 2025 — Mar 2026
  Data Analyst · FPT AdOne — PixOn Game Studio

  Diagnosed player drop-off and ran A/B tests across difficulty, ads and IAP pricing in BigQuery — SQL, Python and dbt-modeled KPI dashboards, cutting drop-rate at key levels ~25% and lifting D1 retention ~12%.

• Aug 2023 — Mar 2025
  Data Analyst · BlueStars LLC

  Automated Python ETL (Selenium, Gmail API) for 1,000+ products and 5,000+ Amazon campaigns, and built star-schema Power BI dashboards tracking ACOS/ROAS/TACoS — driving budget calls that lowered TACoS and lifted ROAS.

• Sep 2019 — Jul 2023
  Marketing Management · National Economics University

  GPA 3.61/4.0 — High Distinction, Top 10 GPA, Master's scholarship.