🩺 DataDoc AI — The Data Health Checkup for Your Business

The problem: Every company runs on spreadsheets, and almost every spreadsheet is quietly broken — duplicate rows, missing values, numbers stored as text, outliers nobody caught. That's hours of manual cleanup before anyone can trust a single chart or dashboard. It's one of the most common, recurring problems across every industry that works with data.

The solution: DataDoc AI diagnoses the damage in seconds and treats it automatically, turning a messy export into a clean, trustworthy dataset — with a plain-English report of exactly what was wrong and what was fixed.

One clear story, five steps: Upload → Diagnose → Treat → Understand → Export.

What it does

1. Diagnose — profiles the dataset (rows, missing cells, duplicates, statistical outliers via IQR, columns stored as the wrong type) and turns that into a single 0–100 Data Health Score.
2. Treat — one click fixes what it found: converts mis-typed columns, drops duplicates, imputes missing values, optionally caps outliers — and shows a before/after comparison.
3. Understand — auto-generates plain-English insights (averages, top categories, strongest correlations) using real computed statistics, not an LLM, so nothing is ever hallucinated.
4. Visualize — an interactive distribution chart, correlation heatmap, and category breakdown.
5. Export — download the cleaned CSV, ready to hand off or load into a BI tool.

Tech stack — and why

Tool	Why
Streamlit	Fastest way to turn a Python data pipeline into a real, clickable product a client can use without touching code.
pandas / numpy	All profiling, cleaning, and stats — kept dependency-light on purpose so the whole pipeline is easy to read and explain line by line.
Plotly	Interactive charts that match the dark diagnostic theme.
IQR (Tukey's fences) for outliers	Doesn't assume a normal distribution, unlike a Z-score cutoff — safer on arbitrary, real-world client data.
Template-based insight generation	Compute the statistic first, then plug it into a sentence template. Same technique production BI tools (e.g. Power BI Insights) use — fast, free, and never wrong.

Run it

pip install -r requirements.txt
streamlit run app.py

The app ships with a realistic messy sample dataset, so it's never empty on first load — just click Run Diagnosis & Treatment in the sidebar. Upload your own CSV/Excel file to try it on real data.

How I'd explain this in 30 seconds

"Businesses lose hours cleaning messy exports before they can even start analyzing them. DataDoc AI automates that: it scores how healthy a dataset is, fixes the common problems automatically, and hands back a clean file plus a plain-English report of what changed — no code required from the client."

Project structure

app.py                 # single-page Streamlit app — the five-step story
modules/
  cleaner.py            # profiling, health score, IQR outliers, cleaning logic
  insights.py            # template-based plain-English insight generation
  theme.py                # design tokens + reusable UI components (hero, KPI card, section header)
sample_data.csv         # bundled messy demo dataset

Design tokens (colors, fonts, spacing) live in one file (theme.py) instead of being scattered through the app — change the brand color once, it updates everywhere.