Trading tools
Monitoring a set of trading experiments
An engineering experiment with a visible state
A partner and I explored a set of trading-bot ideas in 2026. He contributed ideas, while I was the main developer working with AI assistance. Several bots ran over a period of months, with a web interface for monitoring and Telegram notifications for opened and closed trades.
The bots needed more than the rules for an individual action. Each process needed market data, a connection to its provider, a record of its local state and a way for us to see what had happened. I include the work for those integration and operating details.
From provider data to recorded events
The documented Go implementations separate provider clients, configuration, database access, strategy logic, position management and logging. They use PostgreSQL to retain ticks, candles, trades and process state, and run as Linux services.
A typical loop obtains a price observation from the external API, stores it, updates the data used by the experiment and evaluates the appropriate rules. Separate position-management work checks existing positions. The process also reconciles local open-state records with the positions reported by the provider, so its own database does not simply become an unexamined second version of reality.
The two provider connections worked differently. One involved OAuth renewal; the other used signed REST requests. Those differences belong in the integration layer while the wider process still needs predictable configuration, status and diagnostic output.
Monitoring the supporting processes too
The surrounding tooling includes token-refresh work, service checks and structured logs. Scheduled wrappers record whether a support task ran, its exit result and a message. The error collection also brings relevant service-journal errors into a common stream. This provides a way to distinguish a process event, an authentication problem and a service-health issue rather than relying only on the next trade notification.
The web view and Telegram alerts make activity accessible to a person, while the detailed records support later inspection. A notification can tell me that something happened; the logs need to help explain which component produced it.
What I take from the project
This remains an experimental project in the portfolio. The same integration work applies to polling services, scheduled connections and operations dashboards: connect the external API, retain process state and give the people responsible a shared monitoring and event view.