AI-Driven Intelligent Shopping Recommendation And Analytics Platform Using Natural Language Processing

Contemporary e-commerce ecosystems confront a structural personalization deficit: conventional keyword-centric retrieval systems cannot interpret intent-rich natural language queries, while static rule-governed recommendation engines fail to capture the dynamic behavioral signals that reveal genuine user preferences. This paper presents the design, implementation, and evaluation of an AI-Driven Intelligent Shopping Recommendation and Analytics Platform that resolves both deficiencies by integrating a transformer-grounded semantic search engine with a behaviorally adaptive collaborative filtering recommendation module. The semantic search component transforms free-text user queries into 768-dimensional dense vector embeddings via a Gemini-powered embedding pipeline and executes approximate- nearest-neighbor retrieval against a MongoDB Atlas Vector Search index, augmented by LLM-extracted structured filter predicates covering price ceiling, minimum star rating, and color attribute. The recommendation engine assigns differential weights to heterogeneous interaction signals—product views, selection clicks, cart inclusions, and completed transactions—aggregating them into continuously updated per-user preference vectors. A scalable three-tier deployment architecture cleanly partitions the React.js presentation layer, the Flask API processing layer, and the MongoDB persistence layer, enabling independent horizontal scaling of AI inference components. Comprehensive testing confirms semantic search response times within two seconds and recommendation feed generation within three seconds, substantially outperforming traditional keyword-based baselines while operating entirely on open-source infrastructure at a fraction of the cost of commercial personalization services.