I build internal systems for offices and small businesses, and ship them complete — in whatever stack you already run.

Problem

501,335 land title numbers across 26 sub-districts and 352 villages, with their title books and survey documents on shelves. The national system records the data but not where the physical sheet is or who is holding it. Staff spent minutes per lookup and lost track of loans.

What I built

• Binder → slot → side location model that answers "where is title #X" for every number in a binder's range before a single document is entered
• Loan, return, overdue and stock-take workflows with printed receipts and QR binder labels
• Column-mapping importer for inconsistent exports from different offices
• 19 reports, daily backups, one-click updater, Windows Service installer that survives reboots

Result

5,896 binders imported and labeled. Lookup under a minute. Installed on an office PC with no internet, used from phones over office Wi-Fi. Source handed over with a signed license.

• 470,989 titles (94%) checked against the shelves; 46,262 recorded as physically missing, a figure the office never had before
• Zero personal data stored: the importer never reads the 458,940 rows of holder names in the source exports
• 92% of historical loans linked to their documents automatically; 3,007 flagged for a person to review instead of guessed
• Village is taken from the code inside the title number, not the typed name: one village spelled two ways in the source (19,371 vs 4,425 records) still lands in one place

Problem

Small private schools track tuition in notebooks and Excel. They have no IT staff, unreliable internet, and cannot afford cloud subscriptions.

What I built

• One executable, no install, runs on the school PC, shared by several clerks over the LAN
• Payments, receipts with amount-in-words, arrears, WhatsApp reminders, student cards, Excel import/export
• Add-on modules (savings, cash book, attendance, admissions) unlocked by signed license codes
• Automatic daily backups and year-end rollover

Result

Complete product: brochure, narrated demo video, tiered pricing by school level, and a reseller tool that issues license codes and tracks sales, commissions and owner cash flow. Deployable by a non-technical person by double-clicking one file.

Problem

Owners of Indonesian boarding houses (kost) track rooms, tenants and monthly rent in notebooks and WhatsApp chats. Every tenant has a different move-in date, so due dates are scattered and late payments slip through.

What I built

• Room map by floor, check-in with printed rental agreement, check-out with deposit settlement
• Rent invoices issued automatically from each tenant's move-in date (monthly to yearly), with Wi-Fi, parking and late fees
• Due-date reminder page: filter "due in exactly N days" or "within N days", then send one by one from the owner's own WhatsApp; the app records who was reminded
• Add-on modules (cash book, electricity & water meters, bookings, maintenance tickets) unlocked by signed license codes

Result

Owner and caretaker work at the same time from a PC and phones on the kost Wi-Fi, without internet. Shipped with pricing by room count, brochure, narrated demo video and a sales commission scheme.

Problem

Neighbourhood treasurers chase monthly dues house by house. Residents have no way to check their status without asking.

What I built

• One server, many neighbourhoods, file-based storage per tenant
• Reminders sent from the treasurer's own WhatsApp (group announcement with a red/green map image, one-by-one on phone, or batch on laptop via a Chrome extension) — no WhatsApp Business API cost
• Residents check status by house number, no account needed
• 60 API tests, Playwright UI tests, extension tested against a mock WhatsApp Web

Result

Runs on the treasurer's laptop over home Wi-Fi today; same binary deploys to a VPS for public access.

Problem

Most small businesses run on exports and spreadsheets: CRM dumps with duplicate customers, order sheets typed by hand, a new Excel file from every branch every month. Cleaning them by hand is slow, and the mistakes end up in reports.

What I built

• Customer Data Cleaner (Python): normalises names, e-mails, phone numbers to E.164, dates and amounts; removes exact and same-person duplicates; every removed or rejected row is logged with a reason
• Order Sheet Automation (Google Sheets): a Data Tools menu that imports a CSV, validates every row, builds a summary with a chart and e-mails a daily report
• Sales Consolidator (Excel): one button merges monthly sheets or a folder of workbooks into a clean master table, flags bad rows and summarises revenue

Result

Nothing is dropped silently: each input row ends up clean, in a duplicates log, or on an issues list with the reason.

• 200,000 messy rows cleaned in under 4 seconds
• Google Sheets sample: 126 rows in, 115 clean orders, 11 flagged
• Excel sample: 127 rows from 3 months, 120 clean, 7 flagged in 0.2 s

Context

Two years as a data-verification lead on national land-registry digitisation projects across four offices. I built the tools my own team needed.

What I built

• Multi-tab browser wrapper for the registry portal with phone-camera OCR feeding a side-by-side reconciliation view (match / mismatch / suggested fix)
• Hardware-locked licenses with signed keys and a self-hosted update feed

Why it matters

I know what it's like to be the user of back-office software. I design for the clerk, not the demo.