- 🌏 Data Scientist & AI/ML Engineer based in Jakarta, Indonesia
- 🧠 Building systems at the intersection of LLMs, RAG pipelines, and production APIs — fintech and real-world analytics
- 📊 Also go deep on credit risk modeling and applied ML in finance
- 🗣️ NLP research contributor at SEACROWD, building open-source resources for Southeast Asian languages
- 🎓 B.Sc. Mathematics, Universitas Indonesia ('23) — thesis on NLP & Twitter sentiment analysis
- 💼 Actively building my AI/ML portfolio and open to new opportunities — full-time or freelance, Indonesian companies or remote international
LLM & AI
ML & Data Science
Backend & DevOps
Data Engineering
Co-author — ACL 2025 (Long Papers) Crowdsource, Crawl, or Generate? Creating SEA-VL, a Multicultural Vision-Language Dataset for Southeast Asia Contributed to dataset construction for Southeast Asian vision-language research.
| Project | Description | Stack | Links |
|---|---|---|---|
| RAG Evaluation & Optimization System | Automated Ragas evaluation harness with an LLM-as-judge layer and a CI/CD quality gate, applied end-to-end to my DocChat and LLM API Production RAG systems with live GitHub Actions gates | Ragas · LangChain · ChromaDB · Groq · GitHub Actions | — |
| Credit Risk Decision Engine | End-to-end credit risk system on 307K applicants — PD modeling, SHAP explainability, expected loss, portfolio analytics, stress testing, drift monitoring | LightGBM · WOE/IV · SHAP · Streamlit | — |
| Document Intelligence RAG System | Citation-grounded RAG over PDFs, DOCX, PPTX, XLSX, and CSVs — every answer traces to an exact page, slide, or row | FastAPI · Postgres · pgvector · Groq · Langfuse | — |
| Resilient Document Extraction Engine | VLM pipeline extracting structured data from receipts, with a primary/fallback model chain and self-healing schema validation | Groq · Llama Vision · GPT-4o · OpenCV · Pydantic | — |
| text2sql-finetune | QLoRA fine-tune of a sub-3B model for fintech text-to-SQL, evaluated on live SQL execution accuracy | Qwen2.5-Coder · QLoRA · Unsloth | — |
| HR Attrition Analytics | EDA, hypothesis testing, classification, regression, and clustering on employee attrition data | XGBoost · Random Forest · Scikit-learn | — |
| Fintech LLM Analyst | Natural language → SQL → BigQuery. Ask questions about 100K+ Olist e-commerce orders in plain English | LangChain · Groq · BigQuery · Streamlit | Live App |
| LLM API Production | Production RAG API with observability tracing, persistent chat history, and CI/CD deployment | FastAPI · LangChain · Gemini · ChromaDB · Langfuse · Supabase · Railway | Live API |
| AI Agent with Tools | ReAct AI agent that chains web search, math, and date tools to answer multi-step questions | LangChain · LangGraph · Gemini · Tavily · Streamlit | Live Demo |
| DocChat — RAG System | Upload any PDF and ask questions about it with source citations | LangChain · ChromaDB · Gemini · Streamlit | Live Demo |
| Aria — LLM Chatbot with Memory | Conversational AI chatbot with persistent in-session memory | Groq · Llama 3.1 · Streamlit | Live Demo |
| Fintech Segmentation API | REST API serving KMeans segmentation on 278K fintech transactions with CI/CD | FastAPI · Docker · Railway · GitHub Actions | Live API |
| Olist Analytics — dbt + BigQuery | E-commerce data pipeline with staging, intermediate, and mart layers | dbt · BigQuery · SQL | — |
| Stock Trading Robot | LSTM model predicting stock prices 10 days ahead with Buy/Sell/Hold signals | TensorFlow · LSTM · Keras | — |
| Face Recognition | Real-time face recognition system | OpenCV · Python | — |
| Indonesian NLP Resources | Curated NLP resources for Bahasa Indonesia and regional languages | — | — |
"Always learning, always building."