Three projects, told without names.
Our clients stay private. The numbers on this page are real and come from our published case files.
We took over a streaming platform another vendor couldn't finish.
The situation
A media company wanted its own subscription video platform: members, payments, permissions, heavy traffic. The previous vendor stalled halfway. Before we could build forward, we had to work out what was left and what could be trusted.
What we did
- Rebuilt it as a five-part .NET 8 solution: member site, staff back office, API, payment service and shared core, on AWS.
- Moved recurring subscription billing into its own service, so a payment problem can't take the platform down.
- Shipped notifications, watch-later and favourites, release dates, four social sign-in providers and a full staff back office.
- When an edge case double-charged some members, we stopped it the same day, fixed it the next, and refunded everyone on day three.
“When the project stalled, they didn't just clean up the mess. They made the whole platform solid. So we handed them the app as well.”
A bilingual website, an 11-module CMS and an AI assistant. One team, end to end.
The situation
An industrial measurement and control group that represents global instrument brands. They didn't want design in one contract, the build in another, and a chatbot nobody maintains six months later.
What we did
- A design system with a bilingual type scale, so English and Chinese both read well.
- 11 bilingual front-end pages, designed in Figma before a line of code.
- An 11-module CMS: products with manufacturer and datasheets, articles, services, partners, version history and scheduled publishing.
- An AI assistant that answers visitors from the company's own documents (RAG on Postgres and pgvector), with a dashboard of top questions and handoff rate.
“We didn't want it split into pieces. We wanted one team to see it through, from the type system to the CMS to the AI assistant.”
Field reps order in plain language. The system does the rest.
The situation
A distributor selling across Taiwan, mainland China, Hong Kong and Malaysia. Reps are out in the field and can't remember every SKU, and one store can sit on several contracts with different prices.
What we did
- An ordering bot inside the chat app the reps already use. It matches stores and products even with typos or shorthand, and applies the right contract price.
- Discounts below contract price go to a manager for approval automatically.
- Photo and OCR intake for handwritten orders. Low-confidence reads go to a person to check.
- Separate permissions for the back office and the bot, four regional price books, and versioned bundles so past reports stay accurate.
“Our reps type ‘store X, contract A, five H3 pens, 10% off’ and the system understands. That's how we wanted to take orders, not by teaching reps an ERP.”
Got a project like one of these?
Tell us where it's stuck. Thirty minutes is usually enough to know whether we can help.