Beep Call

Failed calls from zero-balance subscribers are lost revenue and a missed connection. The operator needed to detect them from a high-volume stream of call detail records, without notifying the same person twice or overwhelming the SMS gateway.

Call records land in Kafka. A Kafka Streams filter keeps only low-balance failures and removes duplicates inside a time window. A bridge hands each event to RabbitMQ, where SMS and flash-call workers send notifications with retries and a dead-letter queue. Redis idempotency keys stop duplicate sends, and every record is also written to ClickHouse for search and reporting.

• The whole pipeline runs locally with one Docker Compose command, including a mock network gateway that stands in for the operator's SMSC and SIP endpoints.
• Prometheus metrics and Grafana dashboards show ingestion and dispatch rates per stage.
• Maven build runs in GitHub Actions on every push.

• Low-balance failure detection from raw call detail records
• Deduplication windows so a callee is not messaged repeatedly
• SMS and flash-call dispatch with retry and dead-letter queues
• Subscriber guard and blacklist
• Call record search and analytics in ClickHouse
• Operator dashboard for configuration, blacklist and dispatch detail

Multi-module Maven project on Java 21 and Spring Boot: common, subscriber-guard, cdr-filter, notification-dispatch, cdr-sink, query-api and app, plus a mock gateway service.

• Kafka for ingestion and Kafka Streams for stateful filtering
• RabbitMQ for dispatch, where per-message retry and dead-lettering matter more than throughput
• ClickHouse as the analytics store behind the query API
• Next.js operator dashboard reading from the query API