Colab 03: Excel Visualization#

This OpenRath v2.0.0 example turns a fixed sales CSV into an editable Excel workbook with formulas, KPIs, charts, previews, and automated quality checks. It demonstrates how a router can block bad input before any artifact is built.

Open in Google Colab · View the workflow source · View the reference workbook

Input and Output#

Input

Output

sales.csv with 24 typed sales rows across six months and four regions.

output/sales-dashboard.xlsx

The generated workbook contains four worksheets:

Worksheet

Contents

Data

The typed source rows in an Excel table.

Calculations

Visible regional, monthly, gross-profit, and return-rate formulas.

Dashboard

Four KPIs plus regional and monthly charts.

Notes

Purpose, traceability, generation, and security notes.

Workflow#

Node

What happens

profile_data

Counts missing cells and duplicate business keys, then computes the source profile.

quality_route

Sends clean data to planning or routes invalid data to stop_for_quality.

plan_dashboard

Requests validated presentation text while preserving program-computed values.

build_workbook

Builds the four-sheet XLSX and four visual previews.

qa_and_register

Rejects missing previews or Excel formula errors before publishing the artifact.

The LLM supplies only bounded presentation wording. Revenue, orders, return rate, region totals, month totals, KPI formulas, and chart data all come from the CSV and deterministic workbook builder.

Quality Routing#

@router(successors=("plan_dashboard", "stop_for_quality"))
def quality_route(self, state):
    return (
        "plan_dashboard"
        if state["profile"]["quality"] == "pass"
        else "stop_for_quality"
    )

For this example, data quality passes only when rows exist and both missing cells and duplicate business keys equal zero. A blocked dataset never reaches the LLM planning or workbook-build steps.

Run and Verify#

  1. Open the Notebook in Google Colab.

  2. Choose Runtime → Run all.

  3. Enter the DeepSeek API key in the masked prompt.

  4. Review the dashboard preview and download the workbook.

The Notebook verifies:

  • the durable run finishes as SUCCEEDED;

  • the profile contains 24 rows and a pass decision;

  • all four worksheets have visual previews;

  • the formula audit finds no #REF!, #DIV/0!, #VALUE!, #NAME?, or #N/A;

  • five checkpoints are stored;

  • the workbook is registered under an artifact://local/ URI.

Adapt It#

Replace sales.csv while preserving the published column contract, or update the profiler and portable builder together for a new schema. Add new quality rules before quality_route so invalid data remains outside artifact-building steps.