Automatic generate flow conversation with only 1 prompt
Analytics realtime
Understands conversations, automatically hands over to a human agent when needed, and delivers high-accuracy STT and TTS.

Call Wiz — AI Call Center for Vietnam

Inspiration

Over 100 million Vietnamese speakers are underserved by AI call center solutions. Most products focus on English, leaving Vietnamese customers frustrated with robotic responses that don't understand their language, accents, or cultural context.

We set out to build something different: an AI call center that truly speaks Vietnamese — not just translates, but understands and responds naturally like a native speaker.

What We Built

CallWiz is a fully AI-powered call center platform featuring:

No-Code Flow Designer — Create complex call scenarios with just natural language prompts
Natural Vietnamese conversations — Customers speak normally, AI understands and responds fluently
Real-time monitoring — Supervisors watch live calls with instant Vietnamese ↔ English translation
Smart human handoff — AI recognizes its limits and transfers to human agents seamlessly
Automatic form filling — Collected data exports directly to Excel templates

Prompt-to-Flow: Empowering Non-Technical Users

The game-changer of our solution: anyone can create AI call scenarios without writing a single line of code.

A bank operations manager simply types:

"Create a flow for customers reporting lost credit cards. Verify their ID and last 4 digits of card number. If verified, lock the card immediately. If verification fails twice, transfer to human agent."

Qwen3-Max instantly generates a complete conversation flow — with greeting scripts, verification logic, error handling, and escalation rules. The visual editor lets users refine with drag-and-drop, or simply type more prompts like "Add a step to ask for callback number".

Why this matters for banks:

Dozens of scenarios, zero engineering cost — Loan applications, account updates, fraud reports, appointment scheduling
Business teams own their flows — No waiting for IT, no developer bottlenecks
Rapid iteration — Test a new script in minutes, not weeks
Consistency across branches — Same AI, same quality, every call

Batch Calling – Reduce workload with one click

Batch 1

Powered by Qwen & Alibaba Cloud

Our solution is built entirely on the Alibaba Cloud AI ecosystem, leveraging cutting-edge Qwen models and cloud infrastructure.

Qwen3-Max — The Brain

Qwen3-Max via DashScope serves as our core reasoning engine:

Intelligent conversation management — Understands customer intent even with unclear expressions, slang, or regional dialects
Multi-step reasoning — Handles complex requests: identity verification → information lookup → action execution
Tool-calling capability — Extracts structured data from natural speech and fills forms automatically
Flow compilation — Transforms visual flow designs into optimized conversation prompts

Custom Vietnamese STT — Trained to Listen

We integrated a Vietnamese-optimized Speech-to-Text model specifically trained for:

Vietnamese phonetics — Recognizes all 6 tones accurately (sắc, huyền, hỏi, ngã, nặng, ngang)
Regional accents — Understands Northern, Central, and Southern Vietnamese dialects
Real-world audio — Handles background noise, phone quality, and natural speech patterns

PAI-EAS OmniVoice — Trained to Speak

We deployed OmniVoice on Alibaba Cloud PAI-EAS for natural Vietnamese text-to-speech:

Custom voice cloning — Created a natural Vietnamese female voice from just 10 seconds of audio
Tone-perfect pronunciation — Correctly pronounces Vietnamese tones that other TTS engines struggle with
Ultra-low latency — First response in ~200ms, enabling real-time conversation flow
Emotional expression — Adjusts tone for empathy, urgency, or reassurance based on context

Qwen MT Flash — Real-time Translation

Qwen MT Flash enables instant Vietnamese ↔ English translation:

Sub-500ms latency — Supervisors see English translations as customers speak Vietnamese
Context-aware translation — Understands banking terminology and customer service phrases
Streaming output — Translations appear word-by-word for long responses

Alibaba Cloud — The Complete AI Infrastructure

Our entire solution runs on Alibaba Cloud, providing enterprise-grade reliability and seamless integration:

Service	Role in Our Solution
DashScope	Powers Qwen3-Max (conversation) + Qwen MT Flash (translation)
PAI-EAS	Hosts custom OmniVoice TTS model with auto-scaling
ECS	Application servers in Bangkok region (low latency to Vietnam)

Why Alibaba Cloud?

Unified AI ecosystem — Qwen models + PAI deployment + cloud infra work together seamlessly
Southeast Asia presence — Bangkok data center ensures <50ms latency to Vietnam
Enterprise compliance — Meets banking security and data residency requirements
Cost efficiency — Pay-per-use for AI inference, auto-scaling for traffic spikes

Challenges We Faced

Vietnamese Language Complexity

Vietnamese has 6 tones that completely change word meaning. "Ma" can mean ghost, mother, horse, rice seedling, tomb, or "but" depending on tone. Standard AI models struggle with this.
Voice conversations require sub-second response times. Any delay feels unnatural.
Customers speak in slang, abbreviations, and regional dialects that are hard for generic models to understand.

What We Learned

Qwen3-Max excels at Vietnamese — Tool-calling works reliably even with colloquial speech
PAI-EAS is production-ready — Custom model deployment in minutes, scales automatically
The Alibaba Cloud ecosystem is cohesive — DashScope, PAI-EAS, and ECS integrate seamlessly
No-code is the key to adoption — Banks embrace AI when business teams control it

What's Next

More Natural Vietnamese Voices

We plan to train multiple voice personas on PAI-EAS:

Professional female — For banking and formal services
Friendly male — For customer support and casual interactions
Regional accents — Northern, Central, Southern Vietnamese options
Emotional range — Happy, empathetic, apologetic tones for different scenarios

SIP Trunking Integration — Real Phone Calls

Currently our solution works via WebRTC (browser/app). Next phase: direct integration with Vietnam telecom networks using Alibaba Cloud SIP Trunking:

Inbound calls — Customers dial a hotline number, AI answers automatically
Outbound campaigns — AI calls customers for appointment reminders, surveys, payment notifications
PSTN connectivity — Works with any phone, no app installation required
Alibaba Cloud Voice Service — Scalable, reliable, integrated with our existing infrastructure

This transforms our solution from a demo into a production-ready call center replacement.

Tech Stack

Component	Technology
Reasoning & Dialog	Qwen3-Max (DashScope)
Speech-to-Text	Custom Vietnamese STT
Text-to-Speech	OmniVoice (Deployed on PAI-EAS)
Translation	Qwen MT Flash
Infrastructure	Alibaba Cloud ECS
Real-time Audio	LiveKit + WebRTC
Frontend	React + TypeScript
Open-source Libraries	Ant Design, React Flow, Axios, LiveKit