Hybrid search
Runs keyword search (BM25) and vector search (Semantic Search) together, then re-sorts the results precisely with a re-ranker.
Technology
Technology proven
in public service.
We built on the CENTRAS AI platform, refined in public-sector operations, and added the technology events need.
A proven foundation
Built on CENTRAS AI.
CENTRAS NS is not an experiment built from scratch. It runs on the search, grounded-answer, multilingual and security technology of the CENTRAS AI platform, already in use at public institutions. We added only what events need.
QuestionIntent & keyword analysis
Keyword searchBM25
Vector searchSemantic Search
Grounded answerRe-ranker → approved materials only · RAG
Tailored knowledge searchGrounded answersUp-to-date informationConversational contextMultilingual answers, 60+Easy-to-understand answersSecure information infrastructure · PII filterOCR & embedding pipelineOperational data insightsFAQ automationDemand forecastingAI auto-reportsSentiment & intent analysissLLM
NS technology: what we added for events
- Real-time question predictionPatent pendingAnalyzes live inquiries, then automatically ranks and surfaces the questions being asked most right now
- Edition learning algorithmRefines prediction and answers with each edition’s data, and feeds unanswered questions into the next edition’s materials
- Story-based precision measurementPatent pendingPrecise state measurement based on a multidimensional, dynamic-response interface
- Attendee profile & assignment engineIn developmentRecords → likes & dislikes → assignment proposals & conflict warnings → outcomes set the baseline for the next assignment
Track record: Climate Action Opportunity Income “Gihu-i”
1.7M+App sign-ups
8K+AI queries per day, on average
70%Repetitive inquiries automated
40%↑Faster inquiry handling
- Gyeonggi Provincial Council “Sowon AI”AI legislative support
- Climate Action Opportunity Income “Gihu-i”Smart public-inquiry support
- Nam-gu, Gwangju “Namgu-On”Public-inquiry agent with automatic language detection
- Gyeonggi Job Foundation “AI Job Navigator”Tailored recommendations · Policy reports
- Korea Chamber of Commerce and Industry “Digital Leading Company Academy Training Support”Course information
- Jeollanam-do “Tansomoa Tantan-e”App use & climate-action participation guidance
Sources for CENTRAS AI deployments and “Gihu-i” figures: centras.ai/cases · centras.ai/about.
System architecture
CENTRAS NS, on one page.
From platform modules to operating principles, everything that makes up CENTRAS NS in six branches.
Credentials
Backed by certifications
and patents.
Search and grounded answers, document processing, security. All refined in public-sector operations before reaching an event.
- T-5+ Excellent Technology RatingAI data analysis and content generation solutions
- Certified Innovative SME (Inno-Biz)
- Selected as a Kibo-Star Valley companyKorea Technology Finance Corporation (KIBO)
- ISO 9001Quality management systemCertified
Grounded answers · RAG
Generates answers only within approved documents. It never builds an answer from information outside them.
Knowledge asset automation
Document collection → OCR (93%+ accuracy) → chunking → embedding → indexing. A new notice becomes answer material right away.
Answers in 60+ languages
Understands the language of each question and answers in the same language. International attendees ask and get answers in the same channel.
Secure information infrastructure
JWT authentication, RBAC access control and TLS 1.3 encryption. Regular expressions and named-entity recognition (NER) filter out personal data.
Specialized language models & data analysis
Integrates and analyzes operational data with an sLLM tailored to the public sector and techniques based on Palantir AIP.
15Patents held
8Patents pending
In AI, deep learning, natural language understanding and data. Listed below are registered Korean patents with published numbers.
- KR 10-1938212Automatic topic-based document classification system that considers meaning and context
- KR 10-2012404Natural language understanding method using the distribution of answer labels across language analyzers
- KR 10-1715118Deep-learning encoding device and method for document sentiment classification
- KR 10-2245942Sentiment-language inference system using AHP and fuzzy integrals
- KR 10-2023491Method and device for collecting and analyzing text data for association-rule analysis
- KR 10-2267835Method and device for analyzing the attractiveness of a target object using deep learning
- KR 10-1590351Data transmission device based on a distributed network protocol, and its data encryption method
Sources for certifications, patents and technology: centras.ai/about · centras.ai/solutions. Patent counts as tallied by JCG Inc. as of October 28, 2026.