Modernized a Trading App for a Riyadh-Based Brokerage Firm to Reduce Order Processing Delays by 42%

An integration adapter translates between the OMS's native message formats and a clean internal domain model, so legacy behavior does not leak into the new services.

• Order states are modeled as a finite state machine: pending submission, acknowledged, partially filled, filled, cancel requested, cancelled and rejected, with only legal transitions accepted.
• Topics partitioned by order identifier so events for one order are processed in sequence.
• Separate consumer groups for app notifications and audit persistence, so a slow consumer never blocks the other.
• Event identifiers, order references, monotonically increasing sequence numbers, and OMS timestamps on every event, enabling duplicate discard and gap detection
• Dead letter topics with alerting and a controlled replay path.

• The client generates a UUID idempotency key per order intent
• The key, a hash of the request body and the submission state, are written in the same transaction as the order record, protected by a unique constraint
• A retry with a known key returns the stored state, and the same key with a different payload is rejected as a conflict
• Orders that time out toward the OMS enter an unconfirmed state and a reconciliation queue with bounded polling
• The app shows "awaiting confirmation" and disables resubmission until the state resolves

• WebSocket channels with per session subscription management
• Snapshot plus delta protocol with sequence numbers, heartbeat frames and resubscription from the last seen sequence after reconnect
• Redis cache aside reads with short TTLs for reference data and recent quotes
• Server-side backpressure that drops superseded quotes for slow clients
• Execution paths read from the OMS, never from cached quotes

• Composite responses shaped per mobile screen
• Composite indexes and EXPLAIN ANALYZE review of the heaviest queries
• Materialized views for end-of-day position summaries
• Keyset pagination replacing offset pagination for transaction history
• RDS read replicas for reporting queries and explicit cache invalidation on order events

• Correlation identifiers propagated through gateway headers, service logs, Kafka message headers and adapter calls
• OpenTelemetry tracing and structured JSON logs
• Dashboards for p50, p95 and p99 latency, consumer lag, WebSocket connection counts and OMS round trip time
• Append-only audit records for order actions and account changes