Pjgalicia runs AI-driven predictive models that process market data continuously, helping location-independent investors reduce risk and act on evidence rather than guesswork, wherever they happen to be working from.
Explore the Models
Pjgalicia ingests high-volume market data in real time and routes it through backtested protocols before any recommendation reaches a user. The objective is variance reduction: narrowing the range of likely outcomes rather than chasing the widest one. This is calm engineering, not speculation dressed up as insight.
The platform is designed to function without a fixed location, time zone or constant human oversight. Each function below addresses a specific logistical constraint of managing capital while travelling.
Pre-configured thresholds let the platform act on rule-based logic between sessions, so a missed connection or a long flight does not translate into a missed signal.
Infrastructure is distributed to keep response times consistent whether a user is connecting from a co-working space in Lisbon or a hotel network in Singapore.
When exposure moves outside a defined tolerance band, the user is notified immediately, with the underlying data point that triggered the alert.
Every output can be traced back through three stages. None of them are hidden behind a black box.
Structured and unstructured market data is collected from multiple exchanges and economic sources, normalised, and time-stamped for consistency.
Trained models identify recurring structures in volatility, correlation and liquidity, flagging patterns that historically preceded meaningful shifts.
Findings are filtered against the user's stated risk tolerance and portfolio composition, producing a recommendation scaled to their actual exposure.
Rather than relying on client testimonials, Pjgalicia publishes the logic of its own validation process. Each model is run against ten years of recorded volatility, including periods of sharp contraction, to observe how its recommendations would have held up.
Illustrative representation of model drawdown recovery across simulated ten-year backtest windows. Not a forecast of future results.
The backtest exists to refine the model, not to market it. Each run surfaces where a protocol under-performed, which informs the next iteration of variance-reduction tuning.
Past performance, simulated or historical, is used internally as a refinement tool. It does not guarantee future results, and no model removes investment risk entirely.
Data is encrypted in transit and at rest, and retained only for the period required to run and refine the relevant models. Access is logged and restricted to the systems that need it to function.
The platform exposes an API for read access to model outputs and alert configuration, allowing it to sit alongside existing brokerage dashboards rather than replace them.
Response times vary with local connectivity, but the underlying infrastructure is distributed across regions specifically to keep processing delay consistent across time zones.