Problem. The operations team kept reporting “calls auto-disconnecting”. Telephony webhooks arrived in three incompatible schema versions, sometimes twice, sometimes out of order, and business logic was silently overwriting telephony state.

Approach. Analysed real Exotel call data against a single resolution engine, found 13 production bugs, and replaced ad-hoc handling with one direction-, duration- and DB-aware enum plus a two-stage verification pipeline (sync webhook + async V3 API) and a reconciliation cron.

Result. False disconnect reports stopped. Status is eventually accurate even when webhook delivery fails.

Problem. A poll-driven inbound call router occasionally assigned the same call to two agents under load.

Approach. Three layered defenses: an atomic conditional UPDATE agent claim, a guarded assignment write, and a Redis SET NX per-call single-flight lock with cached-response replay. Verified with a harness firing 10 simultaneous claimers against real MySQL.

Result. Zero double assignments. Redis loss degrades to the DB claim instead of failing.