Enterprise Contact Center Projects
Building Scalable AI Contact Center Solutions
Worked at Kakao Enterprise Connect Center, a B2B SaaS division developing the company’s AI Contact Center (AICC) solution. The team focuses on building and operating AI-powered chatbots and voicebots to streamline enterprise customer support. Contributed to a SaaS-based AI Contact Center solution for multiple enterprise clients, including Brandi, Hyundai Department Store, Philip Morris, and Kyobo Life.
Date
TYPE
my roles
Product Manager Intern
Training Data
Conversation Design
Testing
Problem definition
Work overload due to repetitive inquiries
The existing call center system fails to efficiently handle repetitive inquiries, leading to agent overload and response delays.
Reliable and prompt responses
Enterprise-level customers expect accurate information, immediate answers, and a trustworthy tone.
Restructuring Intents for Improved Accuracy
Problem
The existing chatbot had overlapping and inconsistent intents, confusing requests such as “exchange” vs. “return” or grouping broad labels like “product issue.” This reduced response accuracy and caused unnecessary escalations to human agents.
Solution
Mapped intents into major service categories (product, order, membership, program).
Incorporated diverse user utterances into the training dataset.
Monitored and refined overlaps using a Confusion Matrix.
Outcome
Restructured overlapping intents into major service categories
Models for the 4 bots I worked on reached 95% intent-classification accuracy, the internal deployment threshold
Building an entity-based Synonym Dictionary
Problem
In fashion e-commerce, customers describe the same concept with synonyms, loanwords, abbreviations, slang, and regional dialects (e.g., hoodie / hood tee / hoody, XL / X-large / extra large). Without normalization and entity alignment, the model treats them as different tokens - hurting intent/entity recognition, increasing fallbacks, and escalating unnecessarily.
Solution
Collected user expressions from FAQs, community forums, app search logs, and real chat transcripts.
Designed an entity-based Synonym Dictionary, mapping each representative term to its synonyms (incl. slang and dialect variants).
Normalized user utterances by entity matching to ensure consistent recognition.
Result
Dictionaries and training sentences grew to 1,000+ entries
Real customer language, including slang, loanwords, and age-specific speech, mapped to consistent entities
Business Impact
Note
"This project was developed for a Korean company and user base, and the original context was preserved as much as possible. Client names and internal terms are disclosed only within the scope of public information"