Amela
For confidentiality, company names, project names, and system images have been replaced with pseudonyms and mocks.
Contact-Center Operator — Platform Build Story

Run phone reception with AI and live SV monitoring.

VoiceDesk AI is an AI platform for call-center operations that integrates PBX connectivity, AI voice response, SV live monitoring, and multi-tenant isolation into a single SaaS foundation. Four roles — OP, SV, OM, SM — operate from one admin console, serving multiple customers (tenants) simultaneously with full isolation.

"Surging call counts, no real-time monitoring, mixed customer data, delayed transfer-rule updates" — the story of launching a new platform, built in-house from scratch in 12 months, that satisfies all four axes call-center operations require at once.

User roles
4Role
OP / SV / OM / SM
Call monitoring
Real-time
WebSocket live transcription
Customer isolation
Multi-tenant
Cosmos DB tenant isolation
Call reception
AI-led
Transfer to SV when needed
01
CLIENT
About the client

A leading domestic "contact-center operator" at the forefront of contact-center operations.

The contact-center operator is a B2B service that takes on call-center work for multiple client companies. Surging call volume and a rising SV monitoring load were major operational challenges. Automating first-line reception with AI response, and giving SV a way to monitor every call in real time, were urgent needs.

Designed as a multi-tenant SaaS platform. A single foundation runs multiple customers (tenants) while keeping data fully isolated per tenant. Four roles — OP, SV, OM, SM — operate from the same admin console according to their respective permissions.

Operating tenants
MultipleConcurrent customers
Handling roles
4Hierarchy
Call history
CompleteRecord
BUSINESS FLOW
01
STEP
Incoming call
Receives tenant inquiries via PBX
02
STEP
AI first-line response
Voice analysis grasps the customer's intent; AI answers common questions
03
STEP
SV live monitoring
SV monitors all calls in real time via WebSocket
04
STEP
Transfer or complete
Complex cases are transferred to OP/SV; simple cases are completed by AI alone
02
CHALLENGE
Design challenges to solve

The four structural requirements demanded of the platform.

None of these were questions of technology choice; they were business requirements tied directly to operational load, customer isolation, responsiveness, and scalability. A new build from scratch required a design that "satisfies them all at once."

① Handling a call surge with people alone breaks the SLA

If operators are added during peak periods, hiring and training costs surge. Conversely, if wait times grow, customer satisfaction drops .

② If SV cannot monitor in real time, anomalies are found only after the fact

If SV checks call status only after recording, problems are found late. Without live monitoring, instant intervention becomes impossible .

③ Retrofitting tenant isolation leaves residual risk

If a single DB manages multiple customers' data, the risk of information leakage is ever-present. A bolt-on implementation impacts schema, authorization, and audit — all of it.

④ Hard-coding transfer rules makes operations rigid

If conditional branches are hard-coded, changes require a deployment. If rules that differ per tenant cannot be handled, customer growth is constrained .

03
ARCHITECTURE
The architecture we built together

Turning phone reception into a "platform that never stops" across four domains.

We built — from scratch and in-house — an AI call-center platform that supports the operations of the contact-center operator. Four structural requirements — AI response platform, real-time monitoring, multi-tenant isolation, and transfer control — were each designed together with Amela into an architecture that delivers lower operational load, customer isolation, and scalability all at once.

Domain ①
AI response platform
Challenge
Handling a call surge with people alone breaks the SLA.
Solution
PBX connectivity + AI voice analysis automate first-line reception
FastAPI plus a voice-analysis module grasps the customer's intent; AI resolves common questions on its own
Domain ②
Real-time monitoring
Challenge
If SV cannot monitor in real time, anomalies are discovered only after the fact.
Solution
Live transcripts are streamed to the SV dashboard over WebSocket
Call status updates every second via Redis + WebSocket; SV can force a transfer in one click
Domain ③
Multi-tenant isolation
Challenge
Retrofitting tenant isolation later leaves a risk of information leakage.
Solution
Cosmos DB fully isolated per tenant
tenantId is embedded in the JWT token, enforcing a tenant condition on every query at the API layer
Domain ④
Transfer control
Challenge
Hard-coding transfer rules makes operations rigid.
Solution
Per-tenant rules applied instantly from the admin console
The SM role sets transfer conditions in the admin console, applied to production with no deployment
TECH STACK
Frontend
Next.js React Radix UI (shadcn) Tailwind i18n
Backend
FastAPI Python 3.11 JWT authentication W WebSocket
Infra & AI
Cosmos DB Redis Docker Voice-analysis AI
04
IMPACT
The value the platform creates

The contact-center operator's "operational efficiency" accelerates.

This platform continuously delivers business outcomes — not technical metrics — such as reduced call load, SV monitoring efficiency, and faster customer expansion.

① A platform that continuously compresses call load

With AI handling first-line reception, human effort concentrates on complex cases. Even peak periods are met by scaling AI rather than adding staff.

② A design that makes SV intervention as fast as possible

SV can live-monitor every call, intervening instantly with a force-transfer the moment an anomaly is detected.

③ A design that flexibly accommodates customer growth

Thanks to the multi-tenant design, adding a new customer just means registering a tenant in the admin console. No new infrastructure required.

④ An architecture that makes operational changes instant

Transfer-rule changes, NG-word additions, and role-permission adjustments are applied instantly by the SM role from the admin console.

📞
STAKEHOLDER
Operator (OP)
First-line response
Typically
At other organizations, staff burn out re-interviewing from scratch on every call.
At the contact-center operator,
AI has already handled first-line reception, so only complex cases are transferred to OP.
Business impact
Cases handled per OP increased
Focus on high-difficulty cases
👁️
STAKEHOLDER
Supervisor (SV)
Monitoring & quality control
Typically
At other organizations, people often listen to recordings later and intervene after the fact.
At the contact-center operator,
While watching the live transcript over WebSocket, SV can force a transfer instantly when needed.
Business impact
Zero delay in SV intervention
After-the-fact response → real-time intervention
🏢
STAKEHOLDER
Operations Manager (OM)
Tenant operations
Typically
At other organizations, adding a customer tends to require system rework or building a new environment each time.
At the contact-center operator,
Just adding a tenant in the admin console lets operations for a new customer start the same day.
Business impact
Faster customer expansion
Scalability of the multi-tenant platform
⚙️
STAKEHOLDER
System Administrator (SM)
Company-wide operations
Typically
At other organizations, operational-rule changes have to wait for a deployment cycle.
At the contact-center operator,
Transfer rules, NG words, and role permissions can be applied instantly from the admin console.
Business impact
Instant operational changes
Waiting for deployment → instant apply
05
DEMO
Three scenarios

Experience the contact-center operator's platform through a working demo.