🟪 GitHub Repo: https://github.com/0marwa/elysium-chatbot
→ Idea: Develop a chatbot that answers Coran-related questions only, using the ChatGPT API, and that simply doesn’t provide an answer when asked an unrelated question.
| Type of question | Example | Response |
|---|---|---|
| Verse requests | What is ayah 3 of surate al Kursi | Return the verse (only in arabic?) |
| Tafsir request | Explain ayah 3 of sourate al fatiha | Return tafsir (in desired language) |
| i3rab request | What’s the i3rab of surah al fatiha ayah 4 | Return i3rab (in desired language) |
| metadata | how many sajdahs is there in the Quran | Return metadata (count) |
| Invalid questions | Anything unrelated | Neutral refusal: “Sorry, I only respond to Quran-related questions” |
| Any related question | What does Islam say about patience | Intelligent answer cross-referenced with the DB |
• Languages
→ React CS: https://react.dev/
→ React notes (mine): ‣
→ ML py CS: https://www.datacamp.com/cheat-sheet/scikit-learn-cheat-sheet-python-machine-learning
→ py CS global: https://www.kaggle.com/discussions/getting-started/154462
→ HTLM CS: https://developer.mozilla.org/en-US/docs/Web/HTML/Guides/Cheatsheet
• APIs
→ ChatGPT API: https://www.chatbase.co/blog/chatgpt-api
→ Gemini Pro API: https://aistudio.google.com/app/apikey
→ Fine-tuning (later): https://developers.google.com/machine-learning/crash-course/llm/tuning
https://www.geeksforgeeks.org/deep-learning/fine-tuning-large-language-model-llm/
Phase 1 :
• Strict prompt to Gemini Pro’s API :
<aside> 📖
"You are an expert in Moroccan Quran interpretation, specifically the Mohammadi school. You have deep knowledge of:
- Quranic verses and their Mohammadi interpretation
- Hadiths relevant to Moroccan Islamic tradition
- I'rab (grammar analysis) of Quranic verses
- Waqf (pause marks) and Sajda (prostration) positions
- Moroccan Islamic scholars and imams
Please answer the following question based solely on this knowledge. If the question is not related to these Islamic topics, politely decline and explain you only answer Quran-related questions.
Question: [user question]"
</aside>
Phase 2 :
• Integrate dataset and do hybrid LLM + data (RAG) Retrieval Augmented Generation :
User Question → Classify Question → If Relevant : Search Database → Retrieve Relevant Content → LLM + Retrieved Data → Intelligent Answer
Irrelevant : Neutral Refusal
Phase 3 (optional but nice): Fine-tuned model + Database
For real AI-powered repsonses & question filtering