Data & warehouses
Star schemas and pipelines that keep the numbers traceable.
BigQuery · DuckDB · SQL## start here
I build data systems people can rely on.
Arya Nathan Bara. Master’s student in data science, Jakarta. Warehouses and BI dashboards, ML and NLP models, an LLM agent, two IEEE papers, and the habit of checking a result before anyone depends on it.
fig. 1jakarta
## what I do4 areas
BI & dashboards
For people who don’t write SQL and still need the answer.
Looker Studio · StreamlitML & NLP
Fine-tuned transformers, CNN benchmarks, an LLM agent.
PyTorch · Hugging Face · scikit-learnChecking the result
Held-out test sets, significance tests, honest error rates.
McNemar · ROC-AUC · error analysis## about
I’m Arya, a data science master’s student in Jakarta. I like the whole pipeline: how the data is put together, what the models do with it, and whether the result survives a hard look.
That has meant dimensional warehouses on BigQuery, dashboards for people who don’t write SQL, NLP and computer-vision benchmarks, and an LLM agent that checks its own answers. Knowing how the data is built makes me usefully suspicious of what comes out of it.
Outside the code, I led a student organization with a 30-person committee serving 500+ members across six campuses, closed over IDR 100M in corporate sponsorships, and published two IEEE conference papers, first author on one. Explaining the work clearly matters to me as much as doing it.
in brief
Based inJakarta, Indonesia
StudyingM.Sc. Data Science, BINUS
Published2 IEEE papers, first author on one
Led30-person committee, 500+ members
AbroadExchange semester, Korea
Looking forData, BI & analytics roles OPEN
## projects5 rows · result first
| no. | project | what I did | result | status |
|---|---|---|---|---|
| 01 | AskData Text-to-SQL agent that verifies its own answers live demo ↗ | Tag each answer trusted, repaired or abstained, then measure whether the checking earns its keep: 150 golden questions, 5 models, 3 runs, McNemar tests. | −69% confidently wrong 21.8% → 6.7% · p = 1.7e-15 |
VERIFIED |
| 02 | MCB — The Archive 47-page archive for a student organization | Content as JSON per generation, so future boards add a chapter without touching code. Next.js, Tailwind, GSAP. | Live 47 static pages |
SHIPPED |
| 03 | Cloud BI Dashboard Olist warehouse to Looker Studio | Kimball star schema on BigQuery, documented metric definitions and lineage so every number can be traced back. | Live BigQuery · Looker Studio · SQL |
SHIPPED |
| 04 | Stress Detection from Social Media Text Benchmark case study, IJCNN 2022 dataset | Fine-tuned RoBERTa. Dropped the “non-advertisement” Twitter split after finding it still full of ads; Reddit subset only. | 97.2% accuracy 0.997 ROC-AUC |
VERIFIED |
| 05 | Gender Classification Three CNNs on CelebA, transfer learning | VGG-19 vs InceptionV3 vs ResNet50, judged on precision, recall, inference time and confusion matrices. | ~91.5% accuracy InceptionV3 led |
VERIFIED |
## changelog
Started Computer Science at BINUS, fast-track programme.
Competition year: favourite presentation at PIMNAS ke-36, grand finalist at the SoCS Hackathon, SASC mentor scholarship.
Associate System Analyst, BINUS IT Division. First production systems, first real stakeholders.
General President, Modelling Club BINUS. First IEEE paper (ICIMCIS 2024).
Exchange semester in Korea. Master’s in Data Science begins in September.
nextM.Sc. graduation, then a data role.
## contact