ADA — Automated Data Analyst

Upload a CSV or Excel file. Get a dashboard, an executive brief, anomaly flags, a forecast, and answers to plain-English questions — with the calculation shown under every number.

Live demo · Documentation · Roadmap · Contributing

What it does

ADA reads your file, works out which column is the metric, which is the date, and which is the segment, then builds the analysis around that.

• Dashboard — trend, segment breakdown, movement waterfall, segment × period heatmap
• Ask ADA — plain-English questions answered locally with pandas, calculation shown
• Anomaly flags — periods outside a calibrated band, sized so a stable series false-alarms about once in twenty analyses
• Forecast — a guarded baseline that refuses to run on thin history and reports when it was no better than assuming no change
• Evidence and next steps — every finding carries its calculation; recommendations are labelled as interpretation, never as cause
• Downloads — Markdown executive brief and cleaned CSV

Limits: 25 MB per file, 250,000 rows analyzed. Formats: .csv, .xlsx, .xlsm.

Ask a business question. Get the number and its calculation.

Focus on one segment. The whole analysis regroups.

Run it

git clone https://github.com/saineshnakra/automated-data-analyst.git
cd automated-data-analyst
python -m venv .venv
source .venv/bin/activate        # Windows: .venv\Scripts\activate
python -m pip install -r requirements.txt
streamlit run app.py

No API key required. The app opens with a built-in demo dataset.

Does my data leave my machine?

No. Cleaning, schema detection, every chart, and every Ask ADA answer are computed locally with pandas.

An optional AI layer adds two things when you supply a key: a query planner for questions the rules cannot parse, and a strategic narrative. Both receive column names, types, and already-computed evidence. Neither receives your rows or cell values. Model-generated code is never executed.

Full details: Privacy · SECURITY.md

Documentation

For the design story behind the project, read I Built an AI Data Analyst That Tells You When It Hallucinates.

Contributing

Good places to start: a new question shape for Ask ADA, a new deterministic metric, schema-detection fixtures, chart accessibility, adversarial test data.

Read CONTRIBUTING.md, browse the good first issues, or pick something from the roadmap.

Every new recommendation needs a test and the calculation that supports it.

License

MIT