科研SAS：AI 科研统计与可视化分析助手

DMSAS is an AI-driven statistical analysis tool that transforms raw data into interpretable, publishable, and reproducible conclusions quickly. It combines the professional capability of traditional statistical software with the intelligence of large language models, enabling researchers without coding backgrounds to perform high-quality analysis on mobile devices.
Core positioning: Let AI handle complex stats while you keep research judgment. Import CSV, select an analysis goal, describe your question – the system automatically completes data understanding, method selection, hypothesis testing, model calculation, result interpretation, and chart generation. Output is structured as text, tables, and charts, ready for papers or presentations.
Dataset management: Built-in module imports local CSV files, auto-identifies field types, missing values, scale, and basic statistics. Columns labeled as numeric, categorical, temporal, or textual. Card-based list supports preview, rename, delete, and quick entry into analysis workflows. Handles hundreds to tens of thousands of rows stably.
62 specialized analysis agents covering mainstream scenarios:
• Trend & change: line/area charts, trend tests, seasonal decomposition
• Group comparison: t-tests (independent/paired), ANOVA (one/multi-way), non-parametric tests, bar/box plots
• Correlation & regression: Pearson/Spearman/partial/canonical correlation, linear/logistic/polynomial/Poisson regression
• Proportion & composition: pie/donut/stacked bar charts, chi-square/Fisher’s exact test
• Distribution description: mean, median, SD, quartiles, skewness, kurtosis, histogram/density/Q-Q plot
• Cluster & classification: K-means, hierarchical clustering, decision tree, scatter/radar/heatmap
• Multivariate evaluation: PCA, factor analysis, reliability/validity, SEM concept explanation
• Funnel & process: funnel chart, stage conversion rates
• Heatmap & matrix: correlation/frequency/difference matrices
• Radar & multi-dimension: multi-indicator comparison
Structured output: Every agent produces three block types – text (background, method, conclusion), table (statistics, coefficients, p-values, CIs), and chart (whitelist: line, bar, pie, scatter, funnel, heatmap, radar). Fixed schemas ensure stable rendering on iOS.
Advanced settings: Customize alpha, CI level, missing value handling (listwise/pairwise deletion, mean/regression imputation). Method stated explicitly in report.
Flexible LLM config: Users can bring their own API key. No forced binding. Data calls initiated by user, protecting privacy and autonomy.
Export & sharing: Full report as text or images. Charts in vector/high-res format. Bookmark favorite agents or result pages.
Privacy & security: Data stored locally. Only sent to configured API when user initiates analysis. No unauthorized upload or analysis. Clear privacy policy.
Typical use cases:
• Medical: clinical trial baseline comparison, efficacy tests, survival concepts
• Social science: