One event contract
Retention, funnels, churn and experiments share event semantics.
Projects / RetentionOS
Versioned release2026-09Product intelligence platform
RetentionOS
RetentionOS turns product events into decision-ready retention, funnel, churn and experiment evidence within one reproducible platform.
Problem
Retention analysis is often split across dashboards, notebooks and disconnected model experiments, allowing metric definitions and lineage to drift.
Ownership
Designed and implemented the event contract, services, analytical queries, churn workflow, experiment guardrails, interface, tests and cloud delivery.
Primary proof
Live end-to-end workflow, v1.0.0 release and 167 backend tests passed
01 / Overview
System and intended use
A portfolio reference for SaaS and product teams evaluating retention analytics, churn workflows, experimentation and internal product-data tooling.
10KSynthetic users
~238KProduct events
167 passedBackend tests
v1.0.0Release
02 / Architecture
System boundary and flow
System flow04 stages
React workspaceAnalyst workspace
FastAPI servicesTyped API boundary
PostgreSQL + RedisDurable product state
Analytics + ML + experimentsDecision modules
03 / Engineering decisions
Key design choices
SQL before prediction
Inspectable cohorts establish descriptive truth before churn scores.
Governed experiments
Power, assignment, guardrails and decision rules are explicit.
04 / Verification
Claims, evidence and boundaries
| Claim | Inspectable Evidence | Boundary |
|---|---|---|
| The backend has a substantial automated verification suite. | The documented run reports 167 backend tests passed with 4 skipped. | Passing repository tests does not establish customer production scale. |
| The frontend's critical behavior has automated coverage. | The documented run reports 8 frontend tests passed. | This is targeted interface verification, not proof of every browser/device combination. |
| The platform demonstrates retention workflows over a sizeable reproducible dataset. | The supplied project evidence records 10K synthetic users and approximately 238K synthetic events. | The users and events are synthetic and do not represent real customers. |
| The portfolio system has a versioned release. | The repository exposes the supplied v1.0.0 release link. | A release tag does not imply a highly available customer workload. |
05 / Boundaries
Limitations and claim boundaries
- Single-node demo, not a highly available customer workload.
- Users and product events are synthetic.
- Churn scores do not establish causality.
06 / Transferable capability
Related client problems
The architecture demonstrates how to connect governed events, product analytics, predictive signals, experimentation, APIs and delivery without presenting the portfolio system as client work.
Relevant to: SaaS analytics, retention intelligence, churn analysis, experimentation and internal product-data tooling.
Discuss a Similar Project →07 / Technical depth
Engineering appendix
Review implementation, failure modes, delivery and trade-offs
Implementation
- Built typed product-event ingestion, durable storage, cohort and funnel queries, churn scoring and experiment workflows.
- Added frontend views, automated backend and interface tests, container delivery, migrations and operational documentation.
Failure modes and controls
- Schema drift is rejected through validation.
- Risk scores are separated from causal claims.
- Cache loss does not replace the durable record.
Delivery and operations
- Google Compute Engine and Docker Compose
- Caddy-managed HTTPS
- Documented migrations and test commands
Engineering trade-offs
- One shared event contract improves consistency but requires disciplined schema evolution.
- A single-node demo keeps delivery inspectable but does not demonstrate high availability.
Discuss a similar project
Need a similar AI/ML system?
Share the problem, current system, data or model constraints and the delivery environment you have in mind.