دخول
English

الجوهرة د.

Retrieval and evaluation for Arabic language models · undergraduate researcher

علوم الحاسب · كلية علوم الحاسب والمعلومات

المستوى ٨ من ٨

المستوى الأخير · خريف ٢٠٢٦

بعد التخرج

ربيع ٢٠٢٧

مهارات مثبتة بمشروع

  • Python
  • PyTorch
  • Hugging Face
٤٤ مشاريع

حُدّث قبل أسبوعين

المشاريع

Dense retrieval for Arabic regulation questions

Fine-tuned an Arabic dense retriever on 1.2k regulation questions; Recall@5 rose from 0.61 (BM25) to 0.74. Short paper at a student workshop.

We built the question set with two law students, who wrote questions against 40 public regulations and marked the paragraph that answers each one. I fine-tuned a multilingual encoder with hard negatives mined from BM25 and compared it with BM25 and an untuned encoder on a split by regulation, so no document appears in both train and test. Most of the gain came on paraphrased questions; questions that reuse the exact wording still favour BM25, so the final system combines both. Code, splits and the evaluation script are public.

الدور: Lead author · retrieval and evaluation

  • Python
  • PyTorch
  • معالجة اللغات الطبيعية
  • معالجة اللغة العربية
  • التوليد المعزّز بالاسترجاع
  • Hugging Face

التعرف على حروف لغة الإشارة السعودية

قيد العمل

يتعرف على ٣٢ حرفًا من أبجدية الإشارة عبر كاميرا الحاسب بدقة ٩٤٪ على ٦ آلاف صورة جمعناها مع ١٢ متطوعًا.

Each volunteer signed every letter under three lighting setups, and we kept two volunteers out of training entirely to test on unseen hands. The model is a small MobileNet fine-tuned on hand crops from OpenCV. Letters that differ only by movement are still the main source of errors, and that is what we are working on now.

الدور: Model and evaluation

  • Python
  • الرؤية الحاسوبية
  • OpenCV
  • TensorFlow

Arabic LLM eval harness

Runs six open Arabic benchmarks against small open models from one config file; used for baselines in our research group's weekly sessions.

  • Python
  • النماذج اللغوية الكبيرة
  • PyTorch
  • Hugging Face

Connect Four agent (course project)

DQN agent trained by self-play; wins 78% of 500 games against a depth-3 minimax baseline.

  • Python
  • التعلّم المعزّز
  • PyTorch

نبذة

Final-term CS student. For the past year I have worked as an undergraduate researcher on Arabic question answering, mostly on retrieval: finding the right paragraph in long regulation documents before any model writes an answer. Our dense retriever raised Recall@5 from 0.61 to 0.74 over BM25 on 1.2k questions, and the work was accepted as a short paper at a student workshop. I like careful evaluation, small reproducible baselines and writing things down. After I graduate in December, I'm looking for an applied research or ML engineering role on Arabic language products.

طلاب مشابهون

العودة إلى العارض