App Service Contact
Industrial Verdict System

An AI decision layer for inspection, authentication, safety and automation.

The domain changes.
The verdict structure doesn't.

Expert judgment, structured as machine-readable knowledge and deployed at the operational bottleneck.

Verdict = the system's final PASS / HOLD / REJECT decision, with its evidence trace.

Explore Modules → See the Architecture
QUAZAR CORE · INDUSTRIAL VERDICT SYSTEM

Swap the domain-specific models and knowledge.
The core architecture stays fixed.

Learn More
01 · DOMAIN — SWAPPABLE LUXURY GOODS ORGANIC WASTE AIRPORT OPS PORT ASSETS ROAD SAFETY PRODUCT PASSPORT 02 · QUAZAR CORE — VERDICT ARCHITECTURE HELD CONSTANT LAYER 1 PERCEPTION detect · segment SWAPPED PER DOMAIN component models LAYER 2 JUDGMENT multimodal evidence LAYER 3 VERDICT XAI · abstention SWAPPED PER DOMAIN ontology content EPO METHOD + VERDICT ARCHITECTURE — UNCHANGED ACROSS DOMAINS 03 · VERDICT OUT PASS HOLD → HUMAN REJECT 04 · ACTION PLC / actuator C-ITS / alert push Trust record write

System Pipeline Flow

01 · Input Domains

  • Luxury Goods
  • Organic Waste
  • Airport Ops
  • Port Assets
  • Road Safety
  • Product Passport

02 · QUAZAR Core Architecture

Layer 1 Swappable
Perception

detect · segment

Swapped per domain

component models

Layer 2
Judgment

multimodal evidence

Layer 3 Swappable
Verdict

XAI · abstention

Swapped per domain

ontology content

EPO method + verdict architecture -
unchanged across domains

03 · Verdict Out

  • PASS
  • HOLD → HUMAN REVIEW
  • REJECT

04 · Action Protocols

  • PLC / actuator trigger
  • C-ITS / alert push
  • Trust record write
3 Layer Verdict Architecture

Perception → Judgment → Verdict

Three things transfer across domains. Nothing else does.

01Perception

Role
High-speed detection & segmentation
Luxury Authentication
Texture, stitching, hardware recognition
GreenVision
Vision Inspector (contaminant detection)

02Judgment

Role
Multimodal evidence fusion & reasoning
Luxury Authentication
Expert VLM™ identity evidence
GreenVision
Quality Estimator (NIR+VLM)

03Verdict

Role
Ontology-constrained verdict → XAI → action
Luxury Authentication
Rule engine → verdict/hold → appraiser
GreenVision
Orchestrator (OWL+PLC+XAI)
Reusable Model & Knowledge Components: Layer 1–3
Object detector ViT classifier NIR CNN VLM reasoner Expert VLM Ontology + rule engine

4 Stage Native Multimodal VLM Pipeline

Better models or knowledge modules replace one stage, not the entire pipeline.

STEP 1 · YOLOv11 + MoE

Object Detection & Expert Routing

STEP 2 · DeiT-v3 / BEiT-v3 (+NIR CNN)

Fine-Grained Feature Classification

STEP 3 · Qwen2.5-VL (+ Expert VLM)

Multimodal Evidence Reasoning

STEP 4 · OWL + SPARQL (+ Rule Engine)

Knowledge Verification & Verdict Orchestration

Trust Infrastructure

The verdict package (evidence, rationale and rule trace) is what gets recorded, not the raw feed.

INPUT

Physical Evidence

JUDGMENT

QUAZAR Expert VLM + Ontology

PERMANENT RECORD

DID + VC · ZKP

based selective disclosure

TRANSFERABLE CORE: Reusable model components · Ontology engineering methodology · Evidence-to-verdict architecture
EPO (Expert Process Ontology) Methodology

The Bottleneck-First Strategy

Start with the highest-value decision bottleneck, not a full-process AI overhaul.

01

Map the Process
End-to-End

Observe how experts perceive, decide, and act across the full process.

02

Identify the Highest-Value Bottleneck

Locate the decision point where errors, delays, cost, or operational risk concentrate.

03

Encode Expert Knowledge

Structure domain concepts, relationships, decision criteria, exceptions, and rules into an ontology.

04

Assemble, Deploy & Validate the Verdict System

Configure reusable perception, reasoning, and verdict components for the target bottleneck, then validate performance against operational KPIs.

Modules · Expanding Across Domains

One Core. Multiple Domain Modules.

The core architecture stays consistent.
Models, ontology content and operating rules adapt to each domain.

One console framework. Eight domain modules.

QUAZAR CORE
VERDICT CONSOLE
GRN  Sorting line, partner plant
V2X  Night inference output, case E04_002
KT  Trust evidence edge platform
+ NEW MODULE

Propose a New Domain

Bottleneck-specific verdict AI can be applied to your process, too.

Propose →
Deployment Portfolio

Client References

Eight modules built. Here is one of them in the field.

Luxury Detective — Client References

9 references
All powered by Luxury Detective
CASE — LUX.C1PoC
Hyundai Department Store

Department-store luxury floor authenticity verdict PoC.

CASE — LUX.C2In Discussion
Gmarket & Auction

E-commerce platform authenticity-verdict integration.

CASE — LUX.C3Reference
Gugus

Secondhand luxury intake appraisal support.

CASE — LUX.C4Reference
Kangwon Land

On-site item appraisal pilot.

CASE — LUX.C5In Discussion
Designer Exchange

UK/Europe secondhand-luxury commerce integration (NDA, LOI, MOU in place).

CASE — LUX.C6Collaboration
KOIPA

IP-protection industry collaboration.

CASE — LUX.C7Infrastructure Partnership
Oracle

Infrastructure expansion partnership (OCI).

CASE — LUX.C8In Discussion
Lotte Department Store

Solution integration discussions in progress.

CASE — LUX.C9Reference
Naver Shopping

One of our retail-platform references.

FAQ

Frequently Asked Questions

Q1

Why not just fine-tune a foundation
model?

Ontology constrains the verdict.
It abstains rather than guesses.

→ 3-Layer Verdict Architecture
Q2

Does the domain transfer
actually hold?

Four domains, one architecture.
Only parts and ontology were rebuilt.

→ Eight modules, one console
Q3

Where does the training data come
from?

DPP compliance capture is labeled data.
One motion, two outputs.

→ Tagless DPP · MODULE — DPP.01
Company

About Us

Organization, certifications and funding history.

Company Profile

FOUNDED 2020.06 HQ Incheon, Korea MISSION Building the Foundation Model for the Physical World
94.5% verdict accuracy - KOLAS verified
12 IP rights - 4 patents registered,
8 applications pending
Venture Company certification, with an in-house R&D center
Selected as a Global IP Star Company
Direct Production Certification (software)

Milestones

2020.06
  • Company founded
2022
  • 11th Kibo Venture Camp · IP Technology Fast-Track
2023
  • Outstanding Company in Data Innovation · blockchain joint research
2024
  • Seed investment — Changgyeong–Newborn
  • SME Technology Development Grant (Didimdol)
2025
  • TIPS Startup Commercialization · Startup Leap Package
  • Google for Startups Cloud Program
2026
  • Pre-A round closed — Coolidge Corner Investment
  • Global Enterprise Collaboration Program

Organization

GROUP 1 · MODEL ASSETS

Verdict Research

Reusable component models every module
draws from.

GROUP 2 · EPO METHOD

Ontology Engineering

Expert VLM pipeline and
data ontology architecture.

GROUP 2 · EPO METHOD

Domain Verification

Judgment criteria and
Human-in-the-Loop verification.

GROUP 3 · DEPLOYMENT

Trust Infrastructure

DID/ZKP anchoring and
large-scale API infrastructure.

GROUP 3 · DOMAIN PRODUCTS

Strategy & Delivery

Global GTM, public programs and
procurement delivery.

Note: The organization mirrors the stack: build (L1), method (L2), ship (L3).