Sage

The problem

One interface, several kinds of work.

A serious personal assistant needs more than a single prompt. Career research, inbox handling, system changes, and deployment each require different tools, context, and risk boundaries. Sage gives those responsibilities to specialist agents coordinated through one phone-first interface.

Architecture with clear ownership

• OrchestratorA CEO-style agent understands intent, delegates the job, and keeps specialist responsibilities separate.
• Career agentDiscovers roles, removes duplicates, grades fit, and moves opportunities through a PostgreSQL-backed state machine.
• Email agentUses OAuth to read and categorise messages. A structural no-send guard means it cannot send mail autonomously.
• Build bridgeTurns a Telegram brief into a planned, reviewed, and deployed change on Railway.

Constraints shaped the product

Efficiency and operating costs remain design constraints. Lightweight infrastructure, selective model use, and explicit agent boundaries keep the system practical to operate and maintain.

Security received the same treatment. I audited the system, tracked the findings, fixed them, and verified the corrections in the deployed environment.

What this demonstrates

Sage is evidence that I can take an ambiguous idea through product framing, orchestration design, persistence, third-party integrations, safeguards, deployment, efficiency, and live verification.