Multi site e-commerce mobile app with artificial intelligence filters

E-commerce platforms give users access to a wide range of products and save time, yet they still struggle to compare items across different websites. Existing keyword-based search systems depend heavily on manual filters, which limit efficiency and accuracy.This thesis presents an AI-powered, voice-enabled mobile application as a more user-friendly solution to product discovery, offering natural voice queries, intelligent filtering, and multi-site data aggregation. The framework opts for Google's STT API to transcribe requests, Hugging Face's zero-shot classifier (bart-large-mnli) to derive product characteristics, and Playwright to collect the on-the-spot data from e-commerce platforms. A Flutter-based frontend communicates with a Flask backend that handles NLP and retrieval. System performance was tested across transcription accuracy, classification reliability, latency, and user satisfaction. Results showed that the STT module achieved accuracy of 83.3%, with peak values near 90%, while the NLP classifier reached 92% accuracy in identifying user intent and categories. End-to-end latency averaged 2.9–3.7 seconds, meeting real-time usability standards. User evaluations indicated a 35% increase in satisfaction compared with traditional keyword search, confirming that combining voice interfaces with NLP improves the e-commerce search experience. This project is a good example of how one can integrate voice communication, zero-shot NLP, and platform scraping into a mobile environment. Keywords: E-commerce, Voice Search, NLP, AI Filters, Web Scraping