Projects

AI-Based Exoplanet Detection

• Designed and evaluated a machine learning classifier for exoplanet detection using NASA open-source datasets (Kepler, K2, and TESS), distinguishing confirmed planets, candidates, and false positives via transit-feature modeling.
• Developed during NASA Space Apps Challenge 2025 (Global Nominee), in collaboration with BRIN (National Research and Innovation Agency) researchers Emmanuel Sungging Mumpuni and Thomas Djamaludin.

Indonesian Financial Sentiment Classifier

• Adapted a pretrained Indonesian ROBERTa sentiment model to the financial domain using large-scale synthetic and masked data augmentation (100k samples).
• Evaluated the adapted model on domain-specific financial news data, achieving 97.49% accuracy and F1-score; published on Hugging Face.

Reasoning-Enhanced Small LLM

• Investigated parameter-efficient fine-tuning (LoRA) on a 0.5B-parameter language model to improve chain-of-thought reasoning performance under resource constraints.
• Published model on Hugging Face.

Stock Movement Prediction from News

• Designed and evaluated a multimodal learning model integrating financial news embeddings, sentiment signals, and temporal features to study their impact on stock movement prediction.
• Constructed and analyzed an end-to-end pipeline integrating data collection, modeling, and visualization.