Johnpaul10j/Telecom-AI-Assistant-NG.

An AI-powered support agent designed to bridge the information gap during the NCC Harmonized USSD transition in Nigeria. Built with a Retrieval-Augmented Generation (RAG) architecture # 📡 Nigerian Telecom AI Assistant (RAG Pipeline) An AI-powered support agent designed to bridge the information gap during the NCC Harmonized USSD transition in Nigeria. Built with a Retrieval-Augmented Generation (RAG) architecture. ### 🔗 🚀 View Live Demo on Hugging Face Spaces --- ## 📖 Project Overview This project solves a real-world problem: helping 100M+ Nigerian subscribers navigate the 2024/2025 transition to unified USSD codes. By using RAG, the model provides grounded, factual answers based on official NCC standards, significantly reducing LLM hallucinations. ## 🏗️ Technical Architecture The system follows a standard RAG (Retrieval-Augmented Generation) workflow: 1. **Document Loading:** Ingests `faq.txt` using LangChain's `TextLoader`. 2. **Chunking:** Splits text via `RecursiveCharacterTextSplitter` (700 char chunks). 3. **Embedding:** Generates 384-dimensional vectors using `all-MiniLM-L6-v2`. 4. **Vector Store:** Stores embeddings in **FAISS** for efficient similarity search. 5. **Inference:** Retrieves top-k contexts and passes them to **Llama-3.2-3B** via Hugging Face Inference API. ## 🛠️ Installation & Local Setup To run this project locally for development: 1. **Clone the repo:** ```bash git clone github.com. cd Telecom-AI-Assistant-NG. pip install -r requirements.txt HUGGINGFACEHUB_API_TOKEN=your_token_here python app.py ``` ## 🧠 Key Challenges & Solutions **Version Mismatch:** Solved a critical ModuleNotFoundError by implementing a MagicMock monkey-patch to maintain compatibility between gradio 5.0 and huggingface-hub. **Response Precision:** Engineered custom prompt templates to force the model to stay within the 1-3 sentence "Support Agent" persona. ## 👤 Author **Umeh Johnpaul** **LinkedIn:** linkedin.com