Trajan Wealth
An AI assistant that helps wealth-management clients understand their own investments — grounded in the authenticated user's actual portfolio. I owned the architecture and led a team of about ten people across backend, AI, design and mobile to get it into production.
The product
The assistant uses the signed-in user's investment portfolio as context. From there a client can ask about their portfolio metrics, explore potential new investments, or ask about market trends — and get answers that refer to what they actually hold, rather than generic market commentary.
Perplexity supplied the AI capability and Auth0 handled authentication. The architecture around portfolio-data transfer, privacy and security is confidential, which is exactly the sort of detail that made the work interesting.
Leading the delivery
I technically led a multidisciplinary team of roughly ten people spanning backend engineering, AI, product design and mobile development, and owned the architecture across all of it.
What went wrong, and what I did about it
The project completed about six months later than planned. The main cause was delivery friction between my Slovenia-based team and Trajan Wealth's internal US development team — a coordination problem, not a technical one, and the kind that no architecture diagram resolves on its own.
To unblock it I stepped well outside architecture and technical leadership and took over specific implementation and environment work myself:
- Prepared the Auth0 configurations for the mobile clients and the backend system.
- Helped establish the test and staging environments.
- Configured MuleSoft APIs and the connections between systems.
- Configured the relevant MuleSoft policies and access controls.
Outcome
The assistant reached production and remains in use. I would rather present this one with the six-month overrun attached than without it: the useful part of the story is what an architect does when the blocker is on the other side of an organisational boundary.