A trader at a broadcast desk in headphones, a condenser microphone on a boom arm, three monitors of candlestick charts and depth-of-market ladders. — A desk broadcasting to thousands, live through the session.

Financial & Advisory

Software that has to run, not just ship

A subscription product built, launched and still operated six years on. Live market audio, a private community platform, market scanners tuned against a backtesting engine, and automated morning briefings. Around 70 repositories and six years of continuous operation.

The problem

Traders act on information in seconds, so nothing can lag. The platform needed live voice broadcast to thousands of listeners, a community space that off-the-shelf chat could not provide, market data flowing continuously, and subscription access gating all of it, without adding latency anywhere.

What we built

Built a real-time voice broadcast system on Mediasoup WebRTC, proven first as a prototype then taken to production as the “squawk” service traders listen to during market hours
Forked and extended Mattermost into a private trading community, with roughly 25 custom plugins covering trade-post syndication, stock cards, roles and permissions, live user feeds and Discord bridging, all deployed on Kubernetes
Split the platform into services: accounts, sessions, tickers and news, each independently deployable
Wrote custom TradingView indicators including multi-timeframe pivots, minimum-move and risk/reward tooling
Automated a pre-market briefing: market news assembled and distributed before the open without anyone writing it by hand
Specified the market scanners with the desk, then built a backtesting engine so the criteria could be tested against historical data rather than argued about
Iterated the scanners on what the backtests actually showed, which turned a trading opinion into something measurable
Ran load testing against the whole stack before opening it to thousands of members

Result

Live voice infrastructure running continuously since 2020
A community platform owned outright rather than rented from a vendor
Market briefings that publish themselves every trading morning
Scanner criteria that improved against measured results instead of instinct

6 yrs

continuous production since 2020

every

trading day, briefing published unattended

~1.5 hrs

of daily editorial work running without a person

Estimate Reading, filtering and assembling a pre-market briefing, estimated at the time it took when done by hand.

Automation running here

Live voice, broadcast in real timeA WebRTC audio service carrying a live market feed to listeners with the latency budget of a phone call rather than a stream. Built from a prototype proving the transport, then run in production continuously since 2020.
Written briefings that publish themselvesMarket news read, filtered and assembled into a briefing that goes out before the opening bell every trading day. It is the editorial job of a person, running unattended.

See all the automation work →

One hub, four things that must agree with it

Membership hubBuilt from scratch. Everything else asks it who you are.

Real-time audioWebRTC squawk, sub-second

CommunityMattermost, ~25 custom plugins

Market scannersNine, tuned against backtests

Member siteCourses, access, billing

Four services, one source of truth about who a member is and what they have access to. The hard part was never any single piece. It was that all of them had to agree, live, during the session, for six years.

Leads that go cold waiting for a callback

An enquiry worth more than a month of marketing spend lands in a shared inbox on a Friday. Somebody sees it Monday. By the time anyone calls, the person has spoken to a competitor or talked themselves out of it entirely.

Read the case →