DataLens – Local AI Data Analysis Workspace

Most people who need answers from a spreadsheet are not data scientists. Cloud analytics and AI tools mean uploading sensitive business or personal data to someone else's servers, and AI assistants that write and execute code on your data can be unpredictable and hard to trust. Cleaning data by hand also loses track of what was changed and when.

A local-first workspace that automates the routine parts of analysis - type detection, quality checks, dashboards - and keeps every cleaning step versioned. For natural-language questions, the AI only plans and explains: the actual work is done by built-in analysis tools that are validated before they run, so the model never writes or executes code. Statistical insights work even without an AI model.

• Upload CSV or Excel files
• Automatic column-type detection and a data-quality score
• Auto-generated dashboards
• Versioned data cleaning - every cleaning step saves a new version you can track
• Ask Your Data: ask questions in plain English, answered by built-in analysis tools that are checked before they run
• A local Ollama model plans the steps and explains the results, but never writes or runs code
• AI Insights: outliers, correlations and monthly trends using standard statistics - works even without an AI model
• Predictive: trains regression models in the background with a progress bar and a Stop button
• Compares models on held-out data they weren't trained on and shows which inputs matter most (feature importance)
• Every model run is saved
• Reports: save, refresh and export analyses as PDF
• Back up and restore the whole workspace
• Runs entirely on your machine - no sign-in, no telemetry, no Docker required

The backend is a FastAPI service using SQLAlchemy over SQLite to store datasets, versions, model runs and reports, with scikit-learn for statistics and regression modelling; long model-training jobs run in the background and report progress so they can be stopped at any time. The frontend is built with React 19, Vite and Tailwind CSS, with Recharts for dashboards. For Ask Your Data, a local Ollama model turns a question into a plan made of the app's own analysis tools; each step is validated before it runs, and the model then explains the result - it never generates or executes code.

1. The user uploads a CSV or Excel file
2. DataLens detects column types, scores data quality and builds a dashboard
3. The user cleans the data; each step is saved as a new version
4. They ask questions in plain English - the local model plans, validated tools compute, the model explains
5. AI Insights surfaces outliers, correlations and monthly trends
6. They train and compare regression models in the Predictive tab
7. They save analyses as reports and export them as PDF

• Designing safe tool-calling for LLMs
• Local-first, privacy-preserving application design
• Background jobs and progress reporting in FastAPI
• Model evaluation on held-out data and feature importance
• Data versioning and workspace backup/restore