DataPilot: A Self-Hosted AI-Powered Data Analytics Platform with Self-Learning Decision Intelligence

– The explosive growth of enterprise data has created a strong demand for analytics platforms that are simultaneously powerful, affordable, and deployable on-premise. Commercial tools such as Tableau ($70+/user/month) and Microsoft Power BI are constrained by high subscription costs and cloud dependency, while existing open-source alternatives such as Apache Superset and Metabase lack integrated machine learning and conversational analytics. This paper presents DataPilot, a self-hosted, full-stack, AI-powered data analytics platform that combines dashboarding, automated machine learning, natural language querying with multi-turn conversation memory, and a novel self-learning decision intelligence engine within a single application. The system is implemented as a three-service microservice architecture: a Node.js backend with Express and MongoDB; a React 19 frontend with TypeScript; and a Python Flask ML microservice using scikit-learn and XGBoost. A streaming ETL pipeline ingests files up to 5 GB with bounded memory, and a two-level persistent aggregation cache (in-process LRU combined with MongoDB TTL collection) achieves up to 2,060× query performance improvement on a 500,000-row benchmark. Evaluation on five real-world datasets demonstrates a best regression R² of 0.974 using Random Forest, 92% intent-classification accuracy on a 50-query benchmark, and cached response times below 50 milliseconds.