JSON Data Cleaning & Transformation | Demo Project

Synthetic product data transformed with an AI-assisted workflow and independent quality checks. This is a demonstration, not a completed client order.

The source is a small flat JSON product export containing three records. The goal was to produce a consistent active-product structure with explicit field names and numeric quantities, while protecting product identifiers. Success required exact row membership, a declared output schema and verified values and types.

The specification trims surrounding whitespace in label and state, keeps active records and maps code to product_id, label to product_name, state to status and stock to quantity. Quantities are explicitly converted from numeric strings to numbers. Product IDs such as 0101 remain strings; their leading zeros must not be lost.

These operations were executed in an isolated Docker worker. The result was exported as formatted UTF-8 JSON. No additional values were inferred and no external service or production system was updated. The original file, rules and result were retained for independent comparison.

Three source records became two output records after filtering one inactive record. Six text cells were trimmed, four fields were renamed and two retained quantities were converted to numbers. The full comparison image includes every source and result record, making both the retained data and the excluded record inspectable.

All nine validation criteria passed. The separate validator checked the exact schema, retained rows, field values, types, encoding, change accounting, formula safety and SHA-256 output hash. It independently derived the expected result from the source and declared rules rather than accepting the worker's report as proof.

Additional checks compared the result with predefined expected values. A deliberately altered output was rejected. The checks distinguish a string identifier from a number, which matters when an identifier contains leading zeros even though it looks numeric.

The demonstration package contains source JSON, transformed JSON, the explicit transformation specification, a change report, a separate QA report and the complete visual comparison. It illustrates how to prepare and verify a file before an import, without claiming that a live import has occurred.

This is independent work with synthetic data. There was no paying client, client acceptance or measured business uplift. The demonstrated input is flat JSON; nested structures, API extraction, production writes and unsupported formats require separate scope and capability checks. Actual work begins with an agreed field mapping, filtering rule and type policy.