Data Science undergraduate at BINUS University (full scholarship), based in Jakarta. I work on applied machine learning and data products, and I care more about shipping something people can actually open than about notebooks that stay on my laptop.
Currently available for remote contract or part time work in data analysis, dashboards, and applied ML.
Retrieval and LLM systems Hybrid retrieval (dense embeddings + BM25 with reciprocal rank fusion), confidence gating so the model declines instead of hallucinating, and streamed answers with citations. Built on Gemini embeddings and Supabase pgvector.
Data analysis and dashboards Turning messy exports (marketplace CSVs, survey results, spreadsheets nobody wants to touch) into interactive dashboards. Python, pandas, Streamlit, plotly.
Model deployment Getting models out of the notebook: FastAPI services, containerised inference, AWS SageMaker, Hugging Face Spaces.
| Project | What it does | Live |
|---|---|---|
| agentic_verdict | Agentic data analyst with a causal decision engine: sandboxed code execution, self verification, method routing across A/B, observational and time series designs. | open |
| pulse | Real time Jakarta air quality platform. Streaming ingestion, online forecasting with river, drift triggered retraining. | |
| RAG_businessAnalysis_assist | FinSight. Answers questions about a public company strictly from its SEC 10-K filings, with section aware chunking, hybrid retrieval, and a confidence gate that declines instead of guessing. | |
| URL_Detection | PhishGuard v2. URL phishing detection API, MiniLM embeddings into a Keras classifier, with strict input validation and a label orientation regression test. | open |
| nemi-dashboard | E-commerce sales analysis over 20,848 orders, aimed at three owner decisions: which products to grow, which regions are failing, and where margin leaks. | open |
| Feature_shopz | Feature store with a PostgreSQL offline store, a Redis online store, FastAPI serving, and freshness and drift monitoring. | open |
| Addictv2 | CatBoost regressor behind one shared preprocessor so training and serving cannot drift apart. SHAP, tests, CI, Docker. | open |
| iceberg v2 | Scroll driven 3D web CV: a snowfield, a crevasse, then a live-rendered ice cave. React Three Fiber, drei, a world generated from noise. | open |
| web_portofolio_RAG | Ask Nemi. A resume you talk to instead of read: hybrid retrieval over pgvector, confidence gating, streamed answers with citations. | open |
Python pandas NumPy scikit-learn PyTorch CatBoost Streamlit SQL
FastAPI Supabase / pgvector Redis Docker MLflow AWS SageMaker React TypeScript
- LinkedIn: nehemiahwj
- Email: nehewj@gmail.com
Open to conversations about data work, research assistance, and remote internships.


