routing-engine — AI Integration

Status: populated | Last updated: 2026-04-18

Assessment

AI/ML is not applicable to the current routing-engine implementation.

routing-engine is a deterministic, rule-based infrastructure service. Its routing decisions are fully determined by operator-defined rules, health states, and cost/priority metadata. Introducing a model into the hot path (P95 ≤ 50 ms) would add unacceptable latency and non-determinism.

Future Considerations

The following AI-adjacent capabilities may be evaluated in a future roadmap cycle, but are explicitly out of scope for the initial production release:

Capability	Notes
ML-based cost optimisation	A model that predicts delivery rate × cost trade-offs could augment or replace the COST strategy, but requires a reliable DLR feedback loop first.
Anomaly detection on routing outcomes	Detecting degraded operators earlier than the enquire_link heartbeat using delivery rate trends. This would be an offline model feeding operator.health events, not in the request path.
Adaptive strategy selection	Dynamically switching strategy based on time-of-day, operator load, or historical delivery rates. This is a configuration management problem solvable without ML.

None of the above are planned for the current implementation phase.