Bring me
the messy
problem.✳

What was going wrong

SAP reported stock in locations that could not ship orders. Adding it all together made too much stock look available. There were also products sharing a barcode, and warehouses that disappeared from SAP while their old stock stayed visible in OMS.

What I changed

I built and supported two Kafka consumers and a scheduled Python job through QA. The services now count the shippable location, re-read every SKU sharing a barcode before publishing the total, and send zero when a warehouse disappears. I confirmed that last rule with the product owner. An outbox keeps event ingestion separate from the slower work of resolving and publishing stock.

What the QA runs showed

A read-only check of 150 SKUs found 500 stale warehouse positions. The nightly load processed 7,279 SKUs in about 2 min 24 s. The job has 400 tests and 98.8% coverage, with parity tests that replay real QA data to check refactors.

QA results, September–October 2026.

Giving finance an answer

Sales reps need the right credit, and finance needs a reason for every line that does not pay. Territory changes make this harder: the system has to use the rep who owned the territory on the order date. Missing master data also needs to reach someone who can fix it.

The part I owned

I built the allocation job and the Flask API for recalculating a date window, and contributed to Kafka ingestion. The core is a six-step Python rule engine with no I/O. It records each decision, uses territory history, and sends files to Xactly and the finance warehouse over SFTP. Locking and delivery tracking keep recalculations under control.

A QA example

One typical run processed 87 orders in 75 seconds: 75 decisions, 12 exceptions, and zero errors. It delivered 60 commission rows, 87 reporting rows, and a 90-territory history file. The allocation suite has about 300 tests and 88.9% coverage.

QA results, September–October 2026.

Tools for the OMS teams

In Platform Engineering, I led 10 vendor developers in building a shared engineering framework and developer toolbox for L’Oréal’s order management systems (OMS) teams.

Making the QA handoff easier

Writing the code is one part of finishing a story. Someone still needs to check that it does what the business asked for and leave evidence another person can review.

The agents I built

I contributed to L’Oréal’s internal AI-assisted delivery framework. I designed and built QA Analyst and QA Tester agents plus four skills that turn acceptance criteria into use cases, Gherkin tests, and an Excel evidence workbook. Each case points back to a requirement. The execution rules keep runs out of production, redact secrets, and require confirmation before writes.

Where I used the framework

I used the delivery framework across five inventory stories and extended its planning, debugging, and project documentation. The QA role itself is built on a feature branch.