goobolabs/tafsiir-agent

An AI-powered Somali Quran Tafsiir assistant that provides grounded, accurate answers from Somali Quranic interpretation. # Somali Tafsiir Agent > **A high-performance, AI-driven assistant for Somali Quranic Tafsiir.** An AI-powered Somali Quran Tafsiir assistant that provides grounded, accurate answers from Somali Quranic interpretation. Designed for the Somali AI community to make religious knowledge more accessible through state-of-the-art retrieval technology. --- ## 📖 Knowledge Base & Support ### 🧠 Model Memory The intelligence of this agent is anchored by `data/trained.json`. This dataset serves as the primary source for creating the model's semantic memory, ensuring that every response is derived from authenticated Somali Tafsiir. ### 🎯 Supported Scope * **Current AI Agent Focus**: Optimized for **Surah Al-Fatiha** and the **30th Juz (Juz Amma)**. * **System Capability**: While the AI agent is currently tuned for these sections, the underlying **API and architecture support the entire Holy Quran (all 114 Surahs)**. You can easily extend the agent's memory to include the full Quran. --- ## ✨ Key Features - **🎯 Semantic Precision**: Leverages `gemini-embedding-001` to understand deep religious context and Somali nuances. - **⚡ Hybrid Retrieval**: Integrates **Memvid** for a powerful combination of vector similarity and lexical matching, ensuring exact verse lookups. - **🛡️ Hallucination-Free**: strictly grounded in provided tafsiir data. If the information isn't in the context, the agent won't invent it. - **📱 Premium Experience**: A modern, responsive interface with Islamic-inspired aesthetics, smooth micro-animations, and full dark mode support. --- ## 🛠 Tech Stack - **Core**: Next.js 15 (App Router, React 19) - **Styling**: Tailwind CSS 4 (Modern CSS engine) - **Vector Memory**: @memvid/sdk (Local `.mv2` storage) - **Intelligence**: Google Gemini 2.0 Flash - **Data Source**: Custom Somali Tafsiir JSON (`data/trained.json`) --- ## 📁 Project Structure ```bash ├── data/ # Model memory source (trained.json) & Memvid DB ├── src/ │ ├── app/ …