Data science & machine learning — Jakarta, Indonesia

Arya Nathan Bara

I build data and ML projects, then check that they actually work.

Computer science master's student (data science) in Jakarta. I’ve built data warehouses, dashboards, ML models, and an LLM agent, and the part I care most about is the same every time: testing whether the result holds up before anyone relies on it.

View my work Résumé Get in touch
Arya Nathan Bara, smiling while hugging a plush shark LV. 21 · DATA SCIENCE GRADUATE
01

About

I’m Arya. Most of what I make follows one pattern: build the thing, then try hard to prove it wrong. Building AskData’s text-to-SQL agent took a few weeks; deciding how far to trust it took 150 test questions, five models, and a stack of significance tests.

Underneath the ML there’s a lot of ordinary data work I genuinely enjoy: dimensional warehouses, BigQuery, dashboards for people who don’t write SQL, and model benchmarks in NLP and computer vision. Knowing how the data is put together makes me usefully suspicious of what models do with it.

Outside the code, I led a 30+ staff student organization serving 500+ members across six campuses, closed over IDR 100M in corporate sponsorships, and co-authored two IEEE conference papers. Explaining the work clearly matters to me as much as doing it.

02

Projects

Featured · End-to-end analytics + ML + LLM

AskData

“A wrong number that looks right is worse than an error message.”

Text-to-SQL agents are easy to demo and hard to trust: ask a question and the model happily hands you a number, even when it silently answered a slightly different one. AskData is an end-to-end analytics stack on real Olist e-commerce data (~100k orders) — a dimensional warehouse, a review-risk model, and an agent that verifies its own answers, tagging each one trusted, repaired, or abstained instead of guessing. Then I measured whether the verification actually earns its keep: 150 golden questions, five models, three runs each, exact McNemar tests.

The impact

−69% confidently-wrong answers
(21.8% → 6.7%, p = 1.7e-15)
0.756 ROC-AUC
review-risk model
4.1× better than random
at catching bad reviews
  • DuckDB
  • Kimball star schema
  • scikit-learn
  • LLM agent (from scratch)
  • Streamlit
Case study → Live demo ↗ Source ↗
Web · Creative direction

MCB — The Archive

A living digital museum for Modelling Club BINUS: nine generations of history, leadership, and shows across 47 static pages, all driven by plain JSON files so every future board can add their chapter without touching code.

Next.js · TypeScript · GSAP · Lenis Case study →
BI & Analytics

Cloud BI Dashboard

A Kimball star-schema ETL over ~100k Olist orders into BigQuery, plus an interactive Looker Studio dashboard on sales, delivery, and customer-satisfaction KPIs — with documented metric definitions and data lineage for governance.

BigQuery · Looker Studio · SQL Case study →
NLP · Benchmark case study

Stress Detection from Social Media Text

Fine-tuned and benchmarked BERT, RoBERTa, and DistilBERT on the Reddit subset of the IJCNN 2022 stress benchmark; RoBERTa reached 97.2% accuracy (0.997 ROC-AUC). I dropped the dataset’s “non-advertisement” Twitter split after finding it was still full of ads.

PyTorch · Hugging Face · RoBERTa Case study →
Computer Vision

Gender Classification

Compared three CNN architectures (VGG-19, InceptionV3, ResNet50) with ImageNet transfer learning on CelebA; InceptionV3 led at ~91.5% accuracy, judged on precision/recall, inference time, and confusion matrices.

TensorFlow · Keras · CNN Case study →

Published research Co-author on two IEEE conference papers: a TF-IDF sentiment study (EECCIS 2026, in press) and a user-experience comparison of RAG and traditional search (ICIMCIS 2024, DOI). Full list on my résumé →

03

Journey

Started Computer Science at BINUS

Fast-track program, straight into the deep end.

Competition year

Favorite Presentation at PIMNAS ke-36, grand finalist at the SOCS Hackathon, SASC mentor scholarship — and a first taste of presenting technical work to a room.

Associate System Analyst, Bina Nusantara IT Division

SQL projects and system enhancements inside the university's own IT division — first production systems, first real stakeholders.

General President, Modelling Club BINUS

A year leading 30+ staff serving 500+ members across six campuses — closing over IDR 100M in sponsorships, growing Greater Jakarta registrants from 70 to 230, and launching an in-house talent agency for members. Later distilled into MCB — The Archive, the club's digital museum.

First IEEE paper

Co-authored a user-experience study of RAG versus traditional search at ICIMCIS 2024.

Exchange semester in Korea

Artificial Intelligence at Kyung Hee University, plus an award for a cultural-exchange proposal.

Master's in Data Science

GPA 4.00 so far. Built AskData, shipped the Cloud BI warehouse, first-authored the EECCIS 2026 paper, and started the thesis on workplace stress detection.

Expected M.Sc. graduation

Next chapter in progress.

04

Toolkit

Data & BI

  • BigQuery
  • Looker Studio
  • SQL & ETL pipelines
  • Kimball / star-schema modeling
  • DuckDB

Machine Learning & NLP

  • Python (pandas, NumPy)
  • PyTorch · TensorFlow / Keras
  • scikit-learn · XGBoost · LightGBM
  • Transformer fine-tuning (BERT / RoBERTa)
  • Model evaluation

LLM & Delivery

  • LLM agents (text-to-SQL)
  • Evaluation & verification
  • Streamlit
  • Google Cloud Platform
  • Git / GitHub

05 — Contact

Let’s work together.

Open to work and research across data, analytics, and applied ML.

arya.anb.sky@gmail.com LinkedIn ↗ GitHub ↗ Résumé WhatsApp ↗