akeDataAnalyst/ai-interior-material-recommender

AI Interior Material Recommender for Ethiopia: Get ranked matches from 2merkato.com data using sentence embeddings. Built with Python, sentence-transformers & Streamlit. # AI-Powered Interior Material Recommender & Matching Engine An intelligent recommendation system that helps users find matching interior finishing materials (tiles, paints, glass, sanitary ware, etc.) based on natural language queries (style, color, budget, room type, material preferences). ## Project Highlights - Web scraping & data collection from Ethiopian source - con.2merkato.com (Construction Market Watch) - Data cleaning & feature engineering - Text embeddings & similarity search (Hugging Face sentence-transformers) - Interactive recommendation engine - Modern Streamlit dashboard with real-time matching ## Features - Natural language input: "modern blue tiles under 2000 ETB", "white marble for kitchen counter", "frosted glass partition", etc. - Semantic similarity search using sentence embeddings (all-MiniLM-L6-v2) - Ranked recommendations with similarity scores (0–1) - Price formatting (ETB), unit normalization, category filtering - Adjustable parameters: number of results, minimum similarity threshold - Data from real Ethiopian market (2merkato.com) ## Tech Stack | Layer | Tools / Libraries | |-----------------------|------------------------------------------------| | Language | Python 3.10+ | | Data Processing | pandas, numpy | | Embeddings & Similarity | sentence-transformers (all-MiniLM-L6-v2), scikit-learn | | Dashboard | Streamlit | | Data Format | Parquet (embeddings), CSV (backup) | ## Results & Performance - Dataset: 299 real Ethiopian interior/finishing materials (Finishing, Sanitary, Electrical, Tiles & Ceramics, Roofing & Ceiling) - Price range: 15 – 146,500 ETB (mean 3,559 ETB) - Embedding model: all-MiniLM-L6-v2 (384 dimensions) - Similarity performance: Exact/similar items (e.g. Clear Glass variants) score 0.98–0.99 …