Data Employ

A step-by-step guide to the dashboard and a short glossary of every feature.

1 · Load & instant context 2 · Explore & analyze 3 · Clean & preprocess 4 · Train a model 5 · Predict & explain 6 · Monitor in production 7 · Save, share & automate Shortcuts Feature glossary

Workflow 1 · Get your data and see instant context #

Everything starts on the Data tab. Data Employ works with SQLite files, and any SQL database over a URL.

1Add a database. On the left panel, click Upload DB and pick a .db/.sqlite file, or expand Advanced: connect by URL and paste a connection string (e.g. postgresql://user:pass@host/db), then Link.

2Browse your tables — they appear in the left panel with a small sparkline preview of their first numeric column. Click one.

3Quick View launches automatically: a row of stat cards (rows, columns, missing, duplicates), a correlation heatmap, per-column insights, and target analysis — all without any further clicks.

Change the Target column on the Analyze tab, then click ⚡ Quick View to re-render all charts against the new target.

Workflow 2 · Explore & analyze your data #

Open the Analyze tab and pick a type, then click Analyze.

Summary — stat cards, heatmap, insights and target analysis in one view (same as Quick View).
Correlations — a color-coded matrix. Use it to spot multicollinearity before you train.
Column Insights — a distribution band per column (min → max with the mean) and bars of the top values.
Target Analysis — how each feature relates to your target, with correlation bars.
Data Health — a report flagging missing values, duplicates, and near-constant columns.
Export Report (HTML) — download a self-contained, printable HTML report of the current analysis.
Snapshots & Diff — save a result under a name, then diff two snapshots to see what changed between runs.

Workflow 3 · Clean & preprocess #

On the Preprocess tab you build a pipeline of operations, then apply them all at once.

1Pick an operation (e.g. Fill Missing Values), set its params (method=median), and Add Operation. Add as many as you like.

2Click Apply Operations to run the whole queue in order. Watch the result row/column counts.

3Optional: click LLM Suggest to have the advisor propose a cleanup plan automatically.

Operations stay in the queue until you Clear, so you can experiment freely before applying. Export Preprocessed Data (CSV) saves the cleaned result.

Workflow 4 · Train a model #

On the Train tab. If you haven't set a target yet, Auto-Train picks one, prepares the data, and runs in the background.

1Set the target column and a task/tuning mode, or hit Auto-Train to let it decide.

2Watch the run — a progress bar and live log appear; jobs run in the background so the UI stays responsive.

3Read the results — a model comparison bar chart, feature-importance bars, and (for classification) a confusion-matrix heatmap.

4Optional: Check Features (feature health) to drop near-constant, high-cardinality, or highly-correlated columns first.

5Save Model (or Download) so you can reload it later.

The best model becomes the champion automatically. Experiment history is tracked — view it as a leaderboard bar chart on the Train tab.

Workflow 5 · Predict & explain #

Once a model is trained, use the Predict tab.

1Single — paste one JSON record and Predict.

2Batch — pick a table and Predict Table, Predict Loaded Data, or upload a CSV. Every row gets a score.

3Explain — click Explain on any row (or use Explain first prediction row). You get contribution bars: green pushes the prediction up, red pulls it down — so you see why.

4What-If — change one feature on a single row (e.g. years_experience → 15) to see the new prediction. Batch What-If does the same across all rows.

5Export Predictions (CSV) to save the scored table.

Workflow 6 · Monitor a model in production #

On the Schema tab under Model Monitoring / Anomaly Detection.

1Capture Reference on the data you trained on — this is the baseline distribution.

2Later, pick a live table and Check Drift. Features are shown as PSI bars colored stable, moderate, or drifted.

3Re-Score Table scores every row and flags drift plus anomalies automatically.

4For anomalies, use Detect then the interactive scatter — pick any two numeric features to visually spot outliers (red points).

Workflow 7 · Save, share & automate #

Export Report (HTML) / Export Training Report (Markdown) — shareable artifacts you can print or send.
PNG charts — click the ⤓ PNG button on any chart to save it as an image for slides/docs.
Reusable Pipeline Recipes — capture data source + preprocessing + training config so you can reproduce a full run later on updated data (Schema tab).
Snapshots & Diff — version your results and compare (Analyze tab).
Command palette (Ctrl/⌘+K) — jump to or trigger any action by typing; scroll & click to pick.
Persistent state — your active tab, table, target and connection are remembered across reloads.
LLM Advisor — natural-language help for analysis, preprocessing suggestions, and SQL (connect a local LLM in the top-left status).

Shortcuts & tips #

Ctrl/⌘+K — open the command palette.
Loading a table auto-runs Quick View; click it again to refresh against a new target.
Hover any chart for a tooltip; charts can be exported as PNG.
Long-running trainings run in the background — you can keep using the app.

Feature glossary #

Feature	Tab	What it does
Upload DB / connect by URL	Sidebar	Open a SQLite file or any SQL database.
Quick View	Data	Auto-render stats, heatmap, insights, target analysis.
Custom SQL / Validate	Data	Run arbitrary (read-only) queries on your data.
Preprocess queue	Preprocess	Chain ops (fill, drop, scale, encode, sample) and apply together.
Analyze (Summary / Correlations / Insights / Target / Health)	Analyze	Charts and stats for understanding the data.
Snapshots & Diff	Analyze	Save and compare versions of results.
Auto-Train / Train	Train	Fit the best model; show comparison + importance + confusion matrix.
Check Features	Train	Flag near-constant, high-cardinality, correlated columns.
Save / Load / Upload Model	Train	Persist and restore a trained model.
Predict / Predict Table / Batch / CSV	Predict	Score a record, a table, or a file.
Explain / What-If / Batch What-If	Predict	Understand a prediction and test changes.
Re-Score Table / Drift / Anomaly	Schema	Monitor a model against new data.
Recipes	Schema	Reproduce a full pipeline later.
Command palette (⌘K)	Global	Search and run any action quickly.
LLM Advisor	LLM	Natural-language analysis and suggestions.

Data Employ · How-to Guide