yusufmolumo/EduGuide-AI

EduGuide AI is an educational assistant built for African learners, designed to make learning simple, interactive, and accessible. Powered by Awarri’s N ATLaS multilingual AI model, it supports English and major Nigerian languages, helping learners understand concepts in a more local and relatable way. # EduGuide AI – Multilingual Learning & Guidance (Powered by N‑ATLaS) EduGuide AI is a multilingual learning and guidance assistant built on N‑ATLaS (Nigeria’s national open-source multilingual AI model). It explains topics, guides users, and answers questions in English, Hausa, Yoruba, and Igbo with culturally relevant context. The app runs fully on your infrastructure (Kaggle/Colab/local GPU) with self-hosted inference—no external APIs. - N‑ATLaS model: N-ATLaS-LLM - Multilingual… - Attribution: “N‑ATLaS is an initiative of the Federal Ministry of Communications, Innovation and Digital Economy, powered by Awarri Technologies.” --- ## Highlights - Multilingual chat (English, Hausa, Yoruba, Igbo) - Response level control (Beginner, Intermediate, Advanced) - Topic presets and Learning‑path friendly UX - Fast Mode (greedy decoding, token cap) for low latency on T4 - 4‑bit quantization with bitsandbytes for speed and memory efficiency - Prompt trimming and caching to reduce latency - Export chat history (JSON) - Built‑in REST API endpoint: `POST /api/generate` - Single-file app: `eduguide_ai.py` --- ## Solution Overview EduGuide AI is a multilingual learning and guidance assistant purpose‑built for Nigeria and Africa. It helps learners understand complex topics with clear, culturally relevant explanations in English, Hausa, Yoruba, and Igbo. The app is fully self‑hosted: it downloads the N‑ATLaS checkpoint from Hugging Face and serves inference locally on GPU (Kaggle/Colab/local). The UX focuses on speed and clarity (Topic Presets, Response Level control, Fast Mode) so users get helpful answers quickly on modest hardware. Key outcomes: - Local, privacy‑friendly inference with zero external API dependency. - Language‑aware, culturally grounded explanations. - Fast responses on a free T4 GPU using 4‑bit quantization, prompt trimming, and caching. ## Technical Details - Inference stack: PyTorch + Transformers with `AutoModelForCausalLM` and `AutoTokenize …