FelpsXP Platform
The ecosystem’s private core, with two roles: the CORE every module consumes (IAM, database, events, contracts and observability) and the observatory that watches each application, with application security in every layer.
Problem
The ecosystem needed a durable technical foundation before meaningful user-facing surfaces could emerge with consistency and long-term coherence.
Solution
Today: a modular monolith in a monorepo, with runtime support, IAM and organization access v0, a shared observability package and architectural guardrails. Next: the observatory and control centre for every FelpsXP application (Hub, Finance, Sentinel and the modules that follow), with signed per-application event ingestion, security data in retention tiers, dashboards and alerts. Private by design, because it holds raw security data.
Architecture highlights
- Modular monolith with workers: domain rules live in packages, never inside a runtime edge
- HTTP API with no web framework: plain Node covers the current routing surface without an extra dependency
- Shared observability package: structured logging, correlation, and recursive redaction of credential-shaped fields
- Separate HTTP and event contract layers, versioned apart from the rules that implement them
- A single Postgres instance with schema-per-domain separation, rather than a database per domain
Challenges
- Building a shared foundation before a real consuming surface existed to validate decisions against concrete requirements
- Deciding what belongs in the shared Platform versus what should live inside each module
Responsibilities
- Define the modular structure of the monorepo
- Establish runtime, logging, and configuration baselines
- Create architectural documentation and governance direction
- Prepare the technical layer that future surfaces consume
Outcomes
- Its purpose is now settled: CORE and observatory for every application. The Hub became independent of it and will show only anonymous aggregates of what it observes; Sentinel is moving to its own repository
- Three independent copies of the same logger were centralized into one package, covered by automated tests, with a capability none of the three had: redacting sensitive fields before serializing
- The API was designed as an external contract for Finance, a product in a separate repository that will have no access to the platform’s source or database; that integration is not wired up yet
Learnings
- Building infrastructure "as protagonist" first, then deliberately stepping back into an enabling layer once Hub emerged, was the right sequencing - foundation work is easy to over-invest in without a real consumer pulling requirements
- The trigger to centralize had already fired three times before anyone acted: the same logger existed, copied, in three services, and the duplication only became a task once it was measured. Extracting it surfaced a real gap none of the copies handled - not one redacted credentials before writing a log line