QuranAI — AI Powered Islamic Assistant

LLM-based system made using MERN stack providing emotion-aware responses using RAG and curated datasets. LangChain • LLMs • RAG • Python • MERN

About the project

This project is based on a smart and interactive platform where users can explore the Qur’an - Ayahs, Translation & Tafseer - add ayah cards to favorites and ask Islamic questions with meaningful, AI-powered guidance.

Date:

Jul 16, 2025

Services:

Project Details

A fully functional, responsive web app using modern tools, cloud integration, and production-ready features. Developed an AI-powered emotional support chatbot using LLM fine-tuning and RAG, with integration of FAISS-based retrieval. Maintained chat sessions history for each user using JWT authorization and secured authentication system with OTP verification.

Things I Did

Auth System:
Secure Signup/Login with OTP-based password recovery
Chat, History, and Favorites are only available after login for privacy
Ask Anything (Chat):
- Ask any Islamic question — spiritual, emotional, daily-life or ibadah related
- Chat gets saved session-wise for logged-in users
- Responses are enriched with Qur’anic verses, references, and contextual wisdom
- You can copy or download responses easily
Explore Qur’an Section:
- Browse ayahs with translation + Tafsir Ibn-e-Kathir
- Stream Qur’anic recitation in 5 beautiful voices
- Like, copy, and add ayahs to favorites
Dark Mode and Responsive Design for seamless experience on all devices
Favorites Section:
- Your saved ayahs with all original features (Tafsir, recitations, copy/download)
- Only visible to signed-in users for personal reflection
Under the Hood – AI & Retrieval Flow:
The intelligence behind Qur’anAI isn’t just a plug-and-play model. It’s a thoughtfully designed pipeline, built using:
- LLM Orchestration: LangChain, ChatOpenAI, HumanMessage, FastAPI, and Pydantic
- Semantic Search: SentenceTransformer + FAISS + custom context formatting (all-MiniLM-L6-v2)
- Emotion Detection: BERT-based classifier to align emotional tone in responses
- Query Classification: Differentiates between tafsir, general, or spiritual queries for context routing
- Secure API Integration: Using OpenRouter (DeepSeek R1-Qwen3-8B)
- Indexing: Contextually enriched with preprocessed .pkl index for fast retrieval
This pipeline is built to classify user intent, detect emotional undertones, pull the right context, and respond with authentic, meaningful answers.
Tech Stack Used:
Frontend: React (Vite), TailwindCSS, Redux Toolkit
Backend: FastAPI, Node.js, Express.js
DB: MongoDB
AI Models: DeepSeek LLM (OpenRouter), FAISS, Sentence Transformers
Dev Tools: CORS, dotenv, Pydantic, LangChain, indexing, vector storage