Parsifal billing platform

Parsifal's billing operation reconciled supplier feed data by hand. The feeds are messy: ~200K raw rows a month across 108 suppliers, with rows that change or silently vanish between snapshots, feeding invoices for Fortune 500 corporate clients. The system processes about $50K a month, roughly $600K a year annualized.

I was the sole engineer on the platform. I built it end to end in .NET 8 with a WPF front end over SQL Server, with Velopack auto-update so operations staff never touch an installer, and OpenTelemetry plus crash reporting so field failures surface in minutes instead of in angry emails.

The piece I'm most proud of is DriftPass, a diffing engine over successive billing-feed snapshots that detects changed and vanished rows. It surfaced about $25K a month in previously unbilled charges and identified roughly $100K in recoverable underbilling. Fail-closed ledger controls and prior-charge suppression keep a bad feed from becoming a bad invoice.

Over about ten weeks I shipped 26 production releases behind a 505-test regression suite, then wrote the specification, decision log, and data-quality log that let non-technical operators run the system unattended. They still do.

I also built four MCP servers so staff could query and act on internal business data conversationally through AI agents; the Power BI server was adopted in production workflows.

The low point: I wrote a custom obfuscator to protect the shipped IP and it corrupted a production build. I traced the failure to an IL-level bug in the obfuscator and shipped the fix. What I'd do differently: run the obfuscator in CI from day one instead of introducing it mid-stream. A tool that rewrites your binary deserves the same regression gating as the product itself.