I work on warehouse and consignment distribution, and I build the tools that keep its numbers honest: a WMS on Google Sheets and Apps Script, ledger patterns, parsers and dashboards. I own the business rules and the testing; AI writes most of the code.
The system I run at work, end to end, with real screens from the live board.
A plush toy brand's stock lived in five files that never talked to each other. I connected inbound, warehouse slots, shipments, a field app for store visits and partner sell-out into one ledger.
Warehouse records, purchase orders and replenishment. Each one started from a real problem in daily operations.
Stock is never stored as a figure. It is derived from an append-only movement ledger, so a balance can always be traced to its transactions.
Reads barcode and quantity out of partner purchase-order lines, and refuses rather than guesses.
DQLab Excel Hackathon, September 2026: demand profile, min and max levels, reorder decisions and warehouse capacity for a distributor.
DQLab Python Hackathon 2026: which SKUs are rising and which ones sell together.
DQLab SQL Hackathon 2026: spotting procurement records that do not fit their regional pattern.
Distribution and finance case studies. The LPG series uses data simulated from real operating experience at a 3 kg LPG sub-agent in West Sumatra.
Demand against fulfilment, return rate and delivery speed across three areas.
Which simple method forecasts monthly demand per area best, to time allocation and reorders.
Segmenting buyers to prioritise deliveries and catch accounts that are slipping away.
The three LPG studies in one Streamlit app, with an adjustable smoothing factor for the forecast.
Sector performance on the Indonesia Stock Exchange and an LQ45 portfolio optimiser.
I come from the business side, not software engineering. My work sits where the warehouse, the stores and the numbers meet: where stock really is, whether a delivery can be proven, and whether a report can be trusted.
When a process breaks, I write down the rule it should never break again, then build the smallest tool that enforces it, with AI writing most of the code and tests proving each rule holds.