Tashkhees is a professional Arabic medical question classification system.
It integrates deep learning + NLP preprocessing + API intelligence in one pipeline:
? Fine-tuned MARBERT (with LoRA & PEFT) for domain-specific classification.
? Scrubbing & advanced preprocessing: normalization, diacritic removal, tokenization (via CamelTools).
️ Speech-to-text input with SpeechRecognition and text-to-speech feedback using gTTS.
Threshold-based hybrid decision making:
High confidence → classification from fine-tuned model.
Low confidence → fallback to Gemini API.
Unified answers always generated by Gemini, while labels come from the model or Gemini depending on confidence.