FPT Software
Enterprise
Development
Scalability

Semantic Analysis Platform for Call Centre Operations

An NLP and semantic analysis platform for one of Vietnam’s largest IT services firms — turning unstructured call-centre conversations into structured, searchable quality signals.

Region

Vietnam

Duration

Ongoing

Team

6 engineers

Engagement

AI/NLP product build

FPT Software logo

Coverage

Manual sampling → every call

Signals

Topic · Sentiment · Compliance

Languages

Vietnamese + English

The Challenge

What they were up against

FPT Software needed an automated way to analyse call-centre conversations at scale — to surface quality, sentiment, and compliance signals that human QA teams could never cover manually.

  • Manual call QA covered only a small fraction of daily call volume

  • Sentiment, intent, and compliance issues surfaced too late to act on

  • Vietnamese and English mixed within the same conversation

  • No single view of agent performance, call topics, or customer sentiment

Our Approach

How we moved in

We built a semantic analysis pipeline that transcribes, classifies, and scores every call — delivering topic, sentiment, and quality signals to QA and operations teams in near real-time.

  • Discovery with QA, operations, and agent-enablement teams to define signals that drive action

  • Speech-to-text pipeline tuned for Vietnamese and code-switched conversations

  • Semantic classification for topics, intents, and compliance events

  • Streaming architecture designed to scale to full daily call volume

What we built

The solution, in detail

01

Automated transcription and diarisation for every inbound/outbound call

02

Semantic tagging — topic, intent, sentiment, and compliance markers

03

Agent-level analytics and drill-down for QA review

04

Operational dashboards showing trends by topic, agent, and time window

Business Impact

The outcomes that matter

QA coverage expanded from manual sampling to every call analysed

Sentiment and compliance signals surfaced in near real-time

Agent coaching grounded in conversation-level evidence

Foundation for further AI features — summarisation, next-best-action, and call automation

Coverage

Manual sampling → every call

QA analysis coverage

Signals

Topic · Sentiment · Compliance

Extracted per-call signals

Languages

Vietnamese + English

Supported languages including code-switch

Qilin Lab built a semantic layer on top of our call operations that we simply could not have staffed manually. It has changed how our QA and operations teams work.
FPT Software logo

Engineering Leadership

FPT Software

Technology Stack

Built with

Python
PyTorch
Transformers
Kafka
PostgreSQL
AWS

Related Case Studies

Eatverse (Vision Foods) — Agentic-AI Dashboard for a 12-Brand Food Portfolio
Eatverse (Vision Foods)
Enterprise

Agentic-AI Dashboard for a 12-Brand Food Portfolio

Built the operations brain for Eatverse (Vision Foods) — a single dashboard that monitors catalogs and supply chains across 12 brands and thousands of SKUs, with custom AI agents flagging anomalies before they cost money.
Coverage
12 brands · thousands of SKUs
Development
AI / Agents
Spend The Bits — B2B Finance Application & Retail Crypto Wallet
Spend The Bits
Fintech

B2B Finance Application & Retail Crypto Wallet

Rebuilt a payments platform end-to-end and lifted uptime from 80% to 99.9% — turning infrastructure from a liability into a revenue guarantor.
Uptime
80% → 99.9%
Development
DevOps
Scalability
Eduley — Canada's Leading Learning Management System
Eduley
EdTech

Canada's Leading Learning Management System

Powering Canada’s leading LMS with DRM-secured content delivery and Kubernetes-grade scale — thousands of students and educators in concurrent session.
Users
Thousands of active students and teachers
Development
DevOps
Security
Scalability

Have a similar challenge?

Let's discuss how we can deliver the same kind of results for your business.

Get in Touch