I believe data analysis and mining could become the next domain after programming in which AI agents achieve adoption at scale. Based on this idea, I built Xenix with PySide6 — a Windows desktop AI agent for data analysis and mining, designed for non-technical users at small and medium-sized enterprises (SMEs). It aims to enable users to perform data processing, data visualization, and machine-learning-based analysis through natural-language interaction.
Its key features include:
· One-click deployment of core components — including LLMs, embedding models, OCR, and machine learning workloads — to local or remote environments, addressing data security, control, and privacy requirements.
· A retrieval-augmented generation (RAG) knowledge base supporting PDF, PPTX, and DOCX documents.
· A DuckDB-based DSL for inspecting and transforming data, together with Vega-Lite and word-cloud tools for generating visualizations.
· Training and application of more than 30 commonly used machine learning models, covering tasks such as regression, classification, clustering, and recommendation.
This talk will briefly introduce the project’s design, overall architecture, and core technical implementation. A live demonstration will show how Xenix integrates large language models, knowledge bases, data processing, and traditional machine learning capabilities into a unified workflow.