CSV Data Cleaning & Formatting | Demo Project

Synthetic product data. AI-assisted processing with independent quality checks. This demonstration is not a client engagement.

The source is a six-row synthetic inventory export with four columns: sku, item, category and status. The goal was to prepare a consistent active-product CSV while preserving identifiers and product names. Success meant applying only the declared transformations, retaining the four active records and verifying the exported file independently.

I first fixed the schema and export settings: comma delimiter, UTF-8 encoding and LF record endings. All four columns remain strings. The cleanup trims surrounding whitespace only in item and category, then retains rows whose status is exactly active. There is no deduplication, inferred correction or numeric conversion in this example.

The transformation was executed in an isolated Docker worker. The source, transformation specification, resulting CSV and change report were retained. A separate validator derived the expected result from the original source and rules rather than trusting the worker's success message.

Six source rows became four output rows. Twelve text cells were trimmed and two inactive records were filtered out. SKUs 001, 002, 004 and 006 remain strings with their leading zeros intact. The product name Clips, 32 mm retains its comma and is correctly quoted in the CSV. The existing image above shows the entire source and output, not a selected success sample.

All nine validation criteria passed: schema, row count, deduplication behavior, value preservation, types, encoding, change accounting, formula safety and output hash. The deduplication check confirms that no unrequested removal occurred. An additional deliberately altered output was rejected, demonstrating that the checks can detect a corrupted result.

The demonstration package contains source.csv, result.csv, the explicit task specification, a change report, a separate ValidationReport.json and the visual comparison. This illustrates a controlled handoff for a later import: the receiver can inspect the rules, compare the complete data and verify the exact output artifact.

This is an independent demonstration using synthetic data, not paid client work. It establishes the recorded result for this six-row example, not a claimed business improvement or performance benchmark. Real work requires agreed fields, filtering rules and acceptance criteria. Direct production imports, live API writes and unsupported spreadsheet features require separate verification.