Multimodal AI & Medical Vision-Language Models Researcher | Fullstack & Data Science Enthusiast
- 🎓 Undergraduate in Computer Science / Informatics at Universitas Hasanuddin, Makassar, Indonesia.
- 🔬 Current Undergraduate Thesis Research: Selective Prediction for Bilingual Medical Vision-Language Models (VLM) under English–Indonesian Language Shift.
- 💡 Research Interests: Multimodal Foundation Models (VLMs/LLMs), Trustworthy & Safe Medical AI, Selective Classification / Abstention Mechanisms, and Cross-Lingual Robustness.
- 🛠️ Also passionate about modern Fullstack Web Development & scalable systems.
- ⚡ Fun fact: Coffee, anime, and late-night debugging keep the models training!
| 🏥 IndoRad-VQA: Visual Sensitivity–Aware Selective Prediction for Medical VLMs |
|---|
| Investigating diagnostic inconsistency (cross-lingual flip) and text prior hallucination (visual ungroundedness) on radiology VQA when shifting from English to Indonesian. Proposing an inference-time selective risk model combining counterfactual visual deletion sensitivity ($\Delta P_{\text{dir}}$) and cross-lingual disagreement ($D_{\text{lang}}$) to minimize AURC and enable trustworthy clinical referral. |
| Tech: PyTorch • Transformers • Qwen2.5-VL-7B • HuatuoGPT-Vision-7B • Kaggle Dual T4 • Scikit-Learn • IndoRad-VQA |
"Transforming curious ideas into mathematically rigorous models." • Imam D. Qhoir