ANONYMIZED CLIENT / MULTI-TENANT TELEPHONY SAAS / CONTACT CENTERS

Building a multi-tenant telephony SaaS from zero.

I created the shared inbound/outbound voice platform, backend services and AWS operations alone — from SIP edge and ARI call control to tenant configuration, billing, APIs, observability and failure recovery.

ROLESOLE TELEPHONY + BACKEND ENGINEER

THROUGHPUT~100K CALLS / 8 HOURS

PEAK~5K CONCURRENT CALLS

AVAILABILITYNO BUSINESS-HOUR DOWNTIME

THE SITUATION

Many customers needed distinct call products on one shared, high-volume infrastructure.

Banks and other contact-center customers shared the same voice core, but each required its own numbers, dial plans, routing and configuration. New tenants had to become operational immediately while the platform survived extreme bursts and continuous production change.

ARCHITECTURE

Shared infrastructure, tenant-aware behavior and horizontally scaled call control.

Two Kamailio nodes protect the SIP edge. A large Asterisk pool sits behind a clustered ARI proxy; stateless workers execute tenant-specific call flows without coupling application state to one media node.

01 CUSTOMER NUMBERSinbound + outboundtenant identity
→ 02 HA SIP EDGE2× Kamailioactive / passive
→ 03 MEDIA POOLmany Asterisk nodesARI channels
→ 04 CALL CONTROLclustered ARI proxystateless workers
→ 05 TENANT SERVICESRabbitMQ + A2Billingconfig + APIs

WHAT I BUILT / HARD PROBLEMS

One shared platform had to behave like many isolated telephony products.

TENANT-AWARE ROUTING

Numbers, dial plans and configuration selected tenant behavior at the edge, making new-customer provisioning effectively immediate.

HORIZONTAL MEDIA SCALE

Active/passive Kamailio, many Asterisk instances, a clustered ARI proxy and stateless workers absorbed production peaks without sticky application state.

COST + OPERATIONS

AWS decomposition, serverless work and ELK observability reduced cloud cost while keeping failures diagnosable under traffic.

OUTCOME

A shared SaaS platform processing ~100K calls in every eight-hour business day.

Customers could be provisioned almost instantly once commercial setup was complete. Horizontal media capacity, stateless call-control services and production observability kept the platform available through working hours while AWS costs were optimized.

~100K CALLS / 8H / ~5K PEAK CONCURRENT / ~3K SHORT-BURST PEAK CPS / NO BUSINESS-HOUR DOWNTIME

KAMAILIO · ASTERISK ARI · NODE.JS · PYTHON / FLASK · RABBITMQ · A2BILLING · AWS · API GATEWAY / LAMBDA · ELK

Discuss your project