Protocol-First Ingestion
Parquet, JSON, and RSS connectors run on polling intervals with per-source telemetry.
Platform Architecture
One ingestion-to-mobile pipeline for EU safety alerts.
Recall Radar is built as a modular NestJS backend with source-aware connectors, canonical normalization, versioned storage, and mobile-first API delivery.
System Design
Architected for heterogenous public-source data.
Canonical Normalization Layer
Source-specific mappers project records into one ProductAlert contract with provenance.
Idempotent Persistence
Raw payload hashes and version hashes reduce duplicates while preserving record history.
Operational Health Controls
Source states, polling outcomes, and errors are exposed through health endpoints.
Public API Surface
Alerts, filter options, dashboard analytics, GTIN lookup, and device registration.
Mobile Delivery Ready
Expo-based app flow supports feed browsing, dashboard views, and push token onboarding.
Configured Sources
Current live integration scope
- EU Safety Gate historical dataset (Opendatasoft parquet export)
- EU RASFF general information feed (JSON API)
- Germany Lebensmittelwarnung (RSS)
- Italy Ministero della Salute recalls (RSS)
Canonical Data Model
Normalization fields that drive UX
- Canonical alert identity and stable source linkage
- Published and ingested timestamps with normalization context
- Category, risk, country, and source filter dimensions
- Identifiers such as GTIN / lot / model when available
- Raw + canonical version records for traceability
Delivery Lifecycle
Practical path from dev to production
Local
NestJS + Prisma + Docker Postgres for rapid development loops.
Development
GitHub Actions pipeline with staging backend and preview landing builds.
Production
Cloud deployment with health checks, migrations, and monitored source ingest.
Quality Monitoring
Source health is a product surface, not only an ops concern.
Dashboard and API layers both use source reliability signals so teams can judge confidence and freshness in real time.