AI Research Intern @ Institute of Science Tokyo

Building End-to-End
AI & Data Intelligence

Computer Science undergraduate at Universitas Gadjah Mada (UGM) with hands-on experience in AI engineering, temporal event knowledge graphs, LLM corpus engineering, and business data analytics.

Download Resume Explore Work
Rifky Setiawan Profile
Temporal Graph Nets
LLM SFT & PyTorch
Career & Internships

Work Experience

Institute of Science Tokyo (Science Tokyo)

AI Research Intern

Tokyo, Japan Apr 2026 – Present
  • Constructed temporal event knowledge graphs from 50,000 GDELT political events each for the USA and Indonesia via a SQL-Cypher pipeline into Neo4j, applying actor normalization and entity resolution, and identifying a 115–138 day cooperative-to-conflict causal latency across 468,000+ actor interaction sequences.
  • Trained a dual-head Temporal Graph Network (TGN) with gSpan-lite subgraph features for 4-class relation type prediction on the Indonesian political event graph, outperforming XGBoost and logistic regression baselines.
Neo4j Cypher GDELT PyTorch Temporal Graph Networks Python
Ministry of Communications and Digital Affairs (Komdigi)

AI Engineer (Part-time)

Remote, Indonesia Feb 2026 – Jun 2026
  • Engineered a 200M+ token Indonesian language corpus with strict preprocessing (deduplication, toxic text filtering, and character encoding correction) optimized for government and news domains.
  • Produced 20,000 synthetic SFT instruction samples across 4 task types (summarization, issue analysis, ABSA, chatbot) to fine-tune a Qwen3-4B-Base model.
LLM Fine-Tuning Qwen3-4B Corpus Engineering Hugging Face Synthetic SFT
Indonesia Stock Exchange (IDX)

Data Management Intern (Market Development)

Riau Islands, Indonesia Jan 2026 – Mar 2026
  • Built a Medallion data pipeline (Bronze-Silver-Gold) using Python and SQLite, consolidating 6 datasets (604 companies, 55-day OHLCV, 3,300+ IndoBERT-labeled news) into a Star Schema data warehouse.
  • Developed a Power BI dashboard with DAX & SQL views visualizing fundamental-sentiment divergence across 114 liquid stocks for the Sekolah Pasar Modal (SPM) program.
Medallion ETL Power BI DAX SQLite Star Schema IndoBERT
Pertamina Patra Niaga

Data Analyst Intern

Riau Islands, Indonesia Jul 2025 – Aug 2025
  • Developed an interactive Power BI dashboard for Batam's only Green Energy Station (GES) consolidating 500+ daily transactions.
  • Built Python automation scripts processing 3,500+ weekly transactions to surface actionable sales insights to regional management.
Power BI Python Data Analytics Operational Analytics Data Pipelines
AI & Analytics Systems

Selected Projects

🛡️ Multi-Task Learning for Safety-Critical Text

Built a calibrated multi-task model for emotion, GBV, and hate/offense across 30,000+ tweets using a shared encoder and stratified subsets. Applied temperature and isotonic scaling to reduce low-confidence errors, enabling abstention with <10% error at ~90% coverage.

PyTorch NLP Calibration Multi-Task
Colab Notebook

🧾 RAM-Safe Dialogue Summarization with FLAN-T5 Small

Fine-tuned FLAN-T5 Small on 13,460 dialogues (DialogSum) with instruction prefixing and safe batching, achieving ROUGE-1/2/L of 41.3 / 19.7 / 38.2. Maintained factuality with entailment checks and constrained decoding while keeping memory usage under 1.3 GB.

FLAN-T5 Hugging Face Summarization LLM
Colab Notebook

🎧 Audio Sentiment Classification with MFCC and LSTM

Developed a lightweight sentiment classifier using MFCC and LSTM on 2,800 samples from the TESS dataset (7 emotions), achieving macro-F1 optimization. Deployed memory-efficient inference (<25 MB RAM), supporting single & batch audio prediction with CPU latency under 100 ms.

PyTorch Audio Processing MFCC LSTM
Colab Notebook

🚗 Real-Time Car Detection with YOLOv11n

Trained a 1-class vehicle detector on 355 labeled images with YOLOv11n, reaching mAP@0.5 ≈ 0.99 and mAP@[0.5:0.95] ≈ 0.73. Delivered CPU-only inference at ~123 ms/image (≈8 FPS) and ~5.8 FPS end-to-end video processing with annotated MP4 output.

YOLOv11n Computer Vision Object Detection PyTorch
Colab Notebook

🩺 Skin Lesion Classification with GLCM and LBP

Built a classical pipeline with GLCM and LBP features using Logistic Regression, SVM, k-NN, and Decision Tree (best validation accuracy 0.619 with macro-F1 0.62). Ran a stratified evaluation on a 181-image test set with CPU-only inference.

Scikit-learn Computer Vision GLCM & LBP Classical ML
Colab Notebook
Community & Direction

Leadership Experience

UGM KOMATIK Oct 2024 – Oct 2025

Vice Head, Data Mining & AI Division

Komunitas Mahasiswa TIK UGM | Yogyakarta, Indonesia

Led internal selection for GEMASTIK (Data Mining division), reviewing 20+ proposals to secure 10 national delegates, while forging partnerships with 2 external organizations to deliver AI and data mining training workshops.

Batam Campus Expo Aug 2024 – Feb 2025

Founder & General Coordinator

Batam Campus Expo 2025 | Riau Islands, Indonesia

Founded and directed Batam's first large-scale campus expo, leading a 138-member team to connect 2,000+ students across 70 high schools with 32 institutions, securing ~IDR 30M from 10 corporate sponsors.

Awak Gadjah Mada Feb 2024 – Feb 2025

Chairman

Awak Gadjah Mada | Yogyakarta, Indonesia

Steered a 36-member organization to launch a 16-school educational outreach initiative, securing IDR 12M in corporate sponsorships and partnerships with 3 regional government entities.

Academic Background

Education & Achievements

Universitas Gadjah Mada

Yogyakarta, Indonesia | 2023 – Present

B.Sc. in Computer Science

Current GPA: 3.54 / 4.00

Coursework covers algorithms and data structures, probability and statistics, linear algebra, calculus, databases, operating systems, machine learning, and computer networks.

Honors & Awards

Competitions & Recognition
  • 2024
    🏆 Top 10 UGM Delegate (out of 23 teams) Data Mining — GEMASTIK 2024 (Advancing to National Semifinal)
  • 2023
    🎖️ 7th Place out of 65 teams Data Science Competition — GELAR RASA 2023 (UPN Veteran Jawa Timur)
  • 2023
    🥈 1st Runner-Up Contest of Wits — Festival Karakter 2023 (Universitas Gadjah Mada)
Stack & Competencies

Technical Skills

Programming & DB
Python SQL Golang C++ JavaScript BigQuery Neo4j (Cypher)
Data & Analytics
Power BI DAX Excel (VBA) Medallion ETL Star Schema Statistical Analysis
ML & Frameworks
PyTorch TensorFlow Hugging Face Scikit-learn Temporal Graph Net Ultralytics (YOLO)
Tools & Languages
Git Docker Jupyter LaTeX Indonesian (Native) English (DET 120) Japanese (JLPT N5)

Let's Work & Collaborate Together

I am actively open to AI engineering, data science research, and analytics opportunities. Feel free to reach out!

Send Email