gammuush/orochatbot

A localized AI tutor chatbot for Ethiopian elementary school students (grades 4–8) in Afaan Oromo. # 🌐 OroChatbot: Production-Ready Afaan Oromo AI Chatbot A high-fidelity, production-grade conversational AI framework designed specifically for the **Afaan Oromo** language. Built utilizing **FastAPI**, **React SPAs**, **Hugging Face Transformers**, and **Docker**. --- ## 🚀 Key Architectural Strengths - **Afaan Oromo Digraph Awareness**: Tokenizer is fine-tuned to properly digest Afaan Oromo specific digraph segments (like `dh`, `ch`, `sh`, `ny`, `ph`) without sub-tokenizing them incorrectly. - **Morphological Suffix Parsing**: The preprocessing pipeline breaks words down, extracting suffixes (like `oota`, `ummaa`, `dhaan`, `tti`) dynamically to allow morph-semantic similarity matching. - **Dual-Engine Response Pipeline**: - **Primary**: Sequence-to-Sequence (Seq2Seq) text generator using fine-tuned `xlm-roberta-base` encoder-decoders. - **Fallback**: Multilingual semantic similarity retrieval engine. If custom model checkpoints are not yet compiled, the system automatically pools raw question representations and serves appropriate Q&A matches. - **Interactive Controls**: Live training dashboard inside the UI which triggers asynchronous background fine-tuning pipelines and monitors training loss metrics in real-time. - **Complete Validation Suite**: Built-in score calculators computing BLEU-4, ROUGE-1/2/L, accuracy, and model perplexity scores dynamically. --- ## 📁 System Blueprint Structure ``` orochatbot/ ├── backend/ │ ├── app/ │ │ ├── api/ # FastAPI Endpoint Controllers & Limiters │ │ ├── models/ # Schemas & Model Generators │ │ ├── preprocessing/ # Digraph-aware Text Normalizers │ │ ├── training/ # PyTorch Dataset Loaders & Trainer Loop │ │ ├── utils/ # Structured Logger │ │ └── main.py # App bootstrapper │ ├── data/ │ │ └── raw/ # Curated conversational datasets │ ├── Dockerfile │ └── requirements.txt ├── frontend/ │ ├── src/ # React compon …