Internal developer platform and dynamic configuration

Morgan Stanley | Current role

I built a Python, PostgreSQL and React publish/subscribe configuration system that cut update turnaround from two working days to the same day. Admins can update settings without code changes or releases. The system is part of a Python and React internal developer platform for application onboarding, configuration, deployment and operations through shared libraries and tooling.

The configuration problem

Configuration values lived in application code. Updates to AI model names, feature flags, calibration currencies and library-version settings required pull requests, approvals and releases.

What I built

Built a dynamic configuration system with Python, PostgreSQL and React using publish/subscribe. Admins can update AI model names, feature flags, calibration currencies and library-version settings without code changes or releases. Added type validation, dev/QA/prod settings, application/group namespaces and auditing. Cut update turnaround from two working days to the same day.

Applications subscribe to their configuration and react when it changes.

Platform responsibilities

Owned release controls across 4 teams: approval gates, artifact/version tracking, release-status screens and pre-deployment checks.
Owned developer adoption across 4 teams through documentation, sample applications, migrations, Python/React/internal-library upgrades, feedback and feature requests.
Owned support for failed builds, failed releases and slow servers across 4 teams. Contributed to load-balancer health checks and server process startup/restarts.
Owned demo/test environments, promotion between environments, unused application/server cleanup and server right-sizing for one team.

Shared contributions

Contributed to Python and React platform libraries, token validation, service accounts, service discovery and application catalogues with publish metadata and access controls.
Contributed to GitHub Enterprise Cloud onboarding, PR/release jobs, server provisioning, deployment configuration, and scripts and AI skills for server/application creation and deployment.

Outcome

Configuration update turnaround fell from two working days to the same day, removing the need for code changes and application releases for these settings.