Computer Science (Data Analytics) undergraduate at Asia Pacific University, Kuala Lumpur — looking for a data / analytics internship in Malaysia.
I like the full analytics loop: collect → clean → model → visualise → recommend. Everything pinned below is fully reproducible — git clone, one command, and you get the same warehouse, charts, and findings I show here.
Core: Python (pandas, scikit-learn) · SQL (window functions, dimensional modelling) · R (hypothesis testing) Data engineering: DuckDB warehouses, raw → clean → marts pipelines, pytest data-quality suites Visualisation & apps: Streamlit, Altair, Matplotlib, ggplot2 · Next.js/TypeScript when a dashboard needs to be a product Currently exploring: Power BI & BigQuery
| Project | What it shows | Stack |
|---|---|---|
| 🇲🇾 Malaysia Open Data Pipeline | End-to-end ELT pipeline on official DOSM data (CPI + fuel prices): star-schema DuckDB warehouse, 12-check pytest data-quality suite, auto-generated findings, Streamlit dashboard | Python, SQL, DuckDB, Streamlit |
| 🔍 Audit Analytics Toolkit | IT-audit / CAATs journal-entry testing: Benford's law, duplicate payments, approval-threshold splitting, off-hours postings, SoD — every detector proven with pytest against planted ground truth | Python, pandas, pytest |
| 🛰️ ARGUS Intelligence Dashboard | Real-time streaming-data product: live flight/seismic/news/crypto feeds on a 3D globe, anomaly detection, news-driven forecasting, AI daily brief | Next.js, TypeScript, CesiumJS |
| 🛍️ Retail Revenue & Churn | EDA, RFM segmentation, cohort retention, leakage-free churn model (time-split, ROC-AUC ≈ 0.74) | Python, scikit-learn |
| 🗃️ E-commerce SQL Analytics | Pure-SQL cohorts, RFM, and revenue analysis with window functions & CTEs | SQL, SQLite |
| 🧪 A/B Test Analysis | Controlled-experiment analysis: two-proportion z-test, Welch's t, CIs, a clear ship/no-ship call | R, ggplot2 |
Also: I taught myself Swift and shipped Siraj Pro, a prayer-times iOS app with WidgetKit widgets — I enjoy taking things all the way to production.
- Reproducible by default — seeded data generators or committed API snapshots; every README's numbers regenerate from one command
- Tested, not trusted — data-quality checks on pipelines, detectors evaluated against ground truth
- Decision-first — analyses end in a recommendation (ship/no-ship, flagged entries, ranked risks), not just charts
Open to internship opportunities in KL / Selangor — let's talk data.