Yarosoft

Home / Our work

Case study · Trading platform

Software where every decision is checked before it acts.

Few systems are less forgiving than one that places trades on its own. This story is about the engineering, not about trading results.

Trading platform, simplified Data collectors feed a shared host that runs algorithms as plug-in packages. A compute fleet trains and replays models. Every order passes risk checks before execution through broker connections or a Windows bridge. Results publish to a client dashboard with subscriptions and notifications. Data collectors Market and research data Compute fleet Training and replay Shared host Algorithm plug-ins Isolated state per algorithm Risk checks Broker connections and our Windows MT5 bridge Client dashboard Subscriptions, notifications
Simplified system diagram, drawn from the published project details. Not a screenshot.
Challenge
Collect data, decide and act across several machines, without ever losing its place.
What we built
Data collectors, a compute fleet, a shared algorithm host, risk checks, execution and a client dashboard.
Outcome
Every order checked before it acts; restarts pick up exactly where they left off.
Scope
Several machines by role, broker connections and our own Windows MT5 bridge.
Status
Engineering only; no performance figures are published.
Challenge

Why it's hard

Most business software can afford a slow moment or a retry. A trading system can't do the wrong thing twice, or once. Every order must be checked against risk rules first, every result must be reconciled with what the broker actually did, and a restart must pick up exactly where it left off.

What we built

Several parts, working as one

Collectors that gather market and research data, a fleet of machines that trains and tests models, a shared host that runs each algorithm in isolation, a risk manager that checks every order, execution through broker connections, and a client dashboard with subscriptions and notifications.

New algorithms plug in as packages. Adding one needs no change to the platform itself.

Why it matters

Why it matters to you

Your business probably doesn't trade. But if your system needs to take in data, make decisions and act on them automatically, safely and without supervision, this is the standard we build it to.

Automation is only as good as the checks it runs before it acts.

Inside the system

What it looks like in use

Trading workspace showing a currency strength heatmap, market sessions, upcoming events and instrument cards with price charts.
Market data in one place. Currency strength, trading sessions, upcoming events and readings for each instrument. Descriptive readings, not forecasts.From the live system, with identifying details concealed. Select the image to see it full size.
Monitoring cards and gauges for CPU, memory, disk and temperature on each machine; host names and addresses concealed.
The machines behind it. Every machine in the fleet reports its load, memory, disk and temperature, all in one view.From the live system, with identifying details concealed. Select the image to see it full size.
Compared with the usual way

Off the shelf, or built for the business?

A typical approach: an off-the-shelf trading bot or script on a single server, tested by hand on a desktop.

AspectA typical approachWhat we built
Structure One program on one machine doing everything. Separate roles across several machines: data, compute, execution and a client dashboard.
Risk Settings per bot, checked by the bot itself. Account-level risk and certification checks before every new position, outside the strategy.
Changing a strategy Copy a new file over the old one while it runs. Releases pinned to an exact fingerprint; running work finishes before a version changes.
A restart or crash State lost, positions to be checked by hand. Durable journals and reconciliation with the broker; it picks up exactly where it left off.
Research Backtests run by hand on a desktop. Training and replay spread across a fleet of machines.
Brokers One platform, one connection. Broker connections plus a Windows bridge we wrote for MetaTrader 5.
Clients Manual reports and spreadsheets. A dashboard with subscriptions, billing, notifications and live feeds.
Technical detail

For your IT team: what we actually built

A multi-algorithm trading platform split by role across nodes (console, data and compute), a client-facing dashboard, and a Windows execution bridge for MetaTrader 5.

  • Algorithms as plug-in packagesEach declares its own inputs, models, schedules, risk and certification. Publishing, enabling and assigning to bots are separate, deliberate steps.
  • Releases bound to an exact hashContent-addressed releases, kept apart from datasets, models and live state. Activation is pinned to a hash, and running work drains before a version changes.
  • A compute fleetTraining shards and replay jobs run across several machines, separately from the live host.
  • Data collectorsMarket and research sources populated independently of any algorithm's schedule.
  • Risk checked before every entryReadiness and certification checks, account-level risk and execution controls govern every new position.
  • Protective management survives a switch-offA disabled algorithm stops new entries, but positions already open stay protected under the exact release that opened them.
  • Execution across channelsA shared broker outbox with reconciliation, plus a Windows bridge we wrote for MetaTrader 5, with demo and live profiles, durable journals, automatic recovery and protected upgrades.
  • Signed links between partsTrades published to the dashboard are authenticated per source with HMAC signatures; historical replays are labelled and kept apart from real fills.
  • Client dashboardSubscriptions, billing, notifications and execution feeds for clients.

Tell us what's slowing you down.

Describe how things run today, or what worries you. We'll come back with a plain-English view: what we see, what we'd fix first, and roughly what it would involve. It's free, and yours to keep.

Get a free review