I build production web platforms — and stay to run them.

Nothing existed yet — no platform, no infrastructure, no billing. The product had to reach paying customers before it could justify a team, so the architecture had to be something one engineer could build and then operate without it collapsing under support load.

I architected and built it end to end: a Next.js front end over Flask services, trade ingestion from MetaTrader 5, and Stripe plus RevenueCat driving the subscription lifecycle through webhooks. It runs on AWS with a Windows Server tier (IIS, NSSM) for the broker connection, which is Windows-only. I now mentor developers across a ten-person team and still own resolution of infrastructure and customer-facing issues.

Writing the MT5 trade-connection handler in-house

Off-the-shelf MetaTrader 5 packages worked, but they carried a recurring licence cost and put a third party directly in the path of every trade sync — the one thing the product cannot be wrong about. Replacing them meant absorbing the protocol handling and the Windows Server tier it needs, which is real operational surface area I now maintain personally. In exchange the recurring cost went to zero and nothing sits between us and the broker connection.

10,226

Registered users

4,048

Active users

1,944

Paying subscribers

Next.js · Flask · AWS · Windows Server · Stripe · RevenueCat · LLMs

Field-to-Finish coordinates crews and survey orders at scale — millions of rows across multiple Salesflow clients feed its operational and reporting workflows. Its chat, the layer crews actually coordinate through, ran on an external provider the platform didn't control.

Flask and MySQL on Ubuntu, with a Socket.io server and event system I wrote to replace the provider outright. Alongside it I delivered a job-board module that lets field crews bid on available work, and an MCP server that lets AI tools query the platform's data and drive its operations directly.

Owning the realtime layer instead of renting it

The external chat provider worked. But it billed per seat, and every new conversation meant a round trip to their API before anyone could type. Building the socket server meant taking on connection handling, reconnection, and delivery ordering ourselves — work the vendor had been quietly absorbing, and the reason most teams don't do this. It was worth it here because chat is how the crews work, not a feature beside the work.

~50%

Faster chat response and setup

100%

Third-party messaging cost removed

4 yrs

Retained through a restructuring

Python · Flask · Socket.io · MySQL · AWS · Ubuntu · LLMs