Makhfi AI is an intelligent assistant designed for Furqan Qureshi Blogs, a research platform exploring the Quran, Hadith, Modern Science, Archaeology, and Ancient History. It can answer questions, guide users through complex topics, and even summarize videos, based entirely on content from Furqan Qureshi Blogs.
Problem
Furqan Qureshi Blogs needed a way to help users explore complex topics in Quran, Hadith, modern science, archaeology, and history, without manually searching the video content. The goal was to provide reliable, contextual answers and personalized guidance for users.
Solution
Built an AI assistant capable of answering natural-language questions, summarizing videos, and providing context-aware guidance. Implemented a LangGraph agentic RAG pipeline to orchestrate AI reasoning and retrieval, backed by Pinecone for fast semantic search. Long-term memory was integrated via LangMem to allow persistent, personalized interactions.
Outcome
Deployed a production-ready intelligent assistant that allows users to:
- get accurate answers directly from curated content
- summarize any video link from the platform
- maintain context over time with persistent memory
Tech Stack
- Backend & API: Powered by FastAPI for high-performance, asynchronous APIs handling AI interactions, chat sessions, and message persistence.
- Database: Built with SQLModel atop PostgreSQL (via Supabase) to efficiently manage user data, chat history, and metadata.
- Core Agent Framework: Utilizes LangGraph to implement an Agentic RAG pipeline that enables intelligent reasoning and tool orchestration.
- Vector Database: Pinecone enables fast and accurate semantic search across a rich video knowledge base for relevant information retrieval.
- Memory Management: Utilizes LangMem for external, persistent long-term memory—critical for personalized AI behavior.
- Authentication: Uses Supabase Auth for secure, JWT-based user authentication and session management.
