Payroll Preflight

Inspiration

After 13 years of working with payroll and HR systems, I have seen how a small issue—such as a duplicate record, an unusual net payment, or a bank-account conflict—can become a costly problem after payroll is transferred. I wanted to turn that operational experience into a practical tool that helps payroll teams identify risks before money moves.

What it does

Payroll Preflight reviews CSV and Excel payroll files before bank transfer. It detects risks such as duplicate employee IDs, shared bank accounts, net-pay mismatches, non-positive payments, unexplained period variances, post-termination payments, and missing bank-account details.

It produces an explainable HOLD, REVIEW, or READY decision, shows the evidence and recommended action for every finding, and allows users to search and filter the results. Users can also copy a concise review note and export findings or a formatted Excel audit report containing a Summary, Findings Log, and Audit Rules.

How I built it

I built the working prototype with OpenAI Codex and GPT-5.6. Codex helped translate my payroll audit experience into deterministic validation rules, implement the browser-based interface, test the detection logic, and document the project.

Payroll files are processed locally in the browser. The demonstration uses only synthetic data and contains no confidential employee information.

Challenges

The main challenge was converting real payroll-control experience into rules that are strict enough to catch genuine risks while remaining explainable to payroll professionals. Another challenge was supporting both CSV and Excel files and producing an Excel report that can be reviewed, shared, and retained as an audit trail.

Accomplishments

The prototype supports CSV and XLSX uploads, seven payroll audit rules, severity-based filtering, employee search, review-note copying, findings export, and a formatted Excel audit report. The build, lint checks, and all 11 automated tests passed successfully, and the complete payroll audit workflow was manually verified.

What I learned

I learned how domain expertise can be transformed into a working product with Codex, and how important explainability, privacy, and clear follow-up actions are when designing AI-assisted tools for sensitive payroll workflows.

What's next

Next, I would add month-over-month payroll comparison, configurable Saudi payroll controls, approval tracking, annual findings analytics, and secure integrations with HR and payroll platforms such as Oracle.

Built With

• gpt-5.6
• openaicodex