01 · YESTERDAY
“Not what we need.”
✕ON HOLD
✦ FROM THE FIELD TO A GLOBAL STANDARD
Turning field expertise
into data.
Turning data into AI.
We discovered the problems of raw-material trading on the ground: at auctions, in shipments, and in the daily disagreement over quality. We are translating those lessons into EYEBOX and Jeongsongi industrial AI.
EST. 2021SPARKLABS INVESTMENTNVIDIA INCEPTION MEMBER
CORE INTELLIGENCE / JEONGSONGIFIELD DATA ENGINEPERCEPTION → REASONING → ACTIONFIELD DATA → INDUSTRIAL INTELLIGENCE
RAW MATERIALS TO AI · THE FUTURE OF TRADE
SCROLL TO EXPLORE ↓
We traded in the market before we built the technology.
2021Founded
600M KRW2025 B2B distribution sales
26 buyersExisting B2B accounts
27BJEONGSONGI R&D
STARTUP MILESTONES · INVESTMENT & PROGRAMS
Built in the field. Recognized in the startup ecosystem.
A foundation in B2B distribution, strengthened by institutional investment and participation in a global AI startup program.
EQUITY INVESTMENT
SparkLabs seed equity investment · 2023Learn about the investor ↗STARTUP PROGRAM
NVIDIA Inception program member · 2026About NVIDIA Inception ↗
SparkLabs seed equity investment · 2023Learn about the investor ↗STARTUP PROGRAMNVIDIA Inception program member · 2026About NVIDIA Inception ↗◈Field credentials · Garak Wholesale Market intermediary 436
01 / BUYER PAIN, SEEN IN THE FIELD
The same lot. A different decision every day.
Rejected yesterday, approved today. Buyers often lack an objective, reusable quality standard.
“Every order begins with another quality decision.”
SAME PRODUCT · TWO VERDICTSSame commodity photo · different verdicts
Same commodity photo · different verdictsVS
02 · TODAY
“Yes, exactly this.”
✓APPROVED
Unrecorded decisions create repeated time and cost.
✳Repeat sampling✳Repeat negotiation✳Waste and rework risk
Illustrative scenario based on field experiences. Both cards use the same photo from the company’s original IR.
03 / JEONGSONGI — INDUSTRIAL AI CORE
Not an AI for seafood alone. Conversational, code-aware, learning and tool-connected.
From Korean dialogue and developer workflows to feedback-driven adaptation and human-approved business actions, Jeongsongi is designed as a reusable industrial AI agent core.
JEONGSONGIMODEL-AGNOSTIC AI R&D
Continue from our previous task?
✧I can review context and propose the next step.
+ASK JEONGSONGI↗
Conversational AI
Interactive dialogue and context experiments
01def check_order(order):
02 sku = resolve(order.item)
+ require_approval(sku)
+ run_tests()
DIFF VIEW +2 CHANGES
Coding agent R&D
Code analysis, change proposals and testing under development
Field feedback→Train & test→New candidate
LoRASTEP 160
FEEDBACKHOLDOUT
Domain language adaptation
LoRA STEP160 research recorded; evaluation continues
Propose→Review→Approve & run
🛡
HUMAN APPROVAL REQUIREDREAD → PROPOSE → CONFIRMTools with human approval
Proposal, review, approval and controlled execution
27B27B model R&D
LoRA 160160 LoRA steps
VRAM / OOSLong context / VRAM
HUMAN GATEIndependent evaluation
NVIDIA / GPU RESEARCH
Real customers and working prototypes. Compute is the next bottleneck.
Reliable conversation, coding workflows, learning and tools require repeated long-context, concurrent and multimodal evaluation on adequate GPU infrastructure.
INFERENCEFINE-TUNINGPARALLEL EVALMULTIMODAL
02 / ONE CONNECTED BUYING FLOW
From an auction-floor photo to a buyer-approved order.
Connecting existing applications, transaction history and prototypes into one quality-to-order journey.
Auction capture
Capture actual lots
AI quality analysis
Size · color · mix
BUYER HISTORY
SIZE
COLOR
PRICE
Historical buyer data
Preference · price
LOT
BUYER A ✓BUYER B ✓BUYER C ○
Buyer matching
Recommend fit
BUYER REPORTQUALITY + PRICE↗
Tailored offer
Photo · quality · value
ORDER REVIEW✓PRICE · QTY · STOCKCONFIRM ORDER
Approve & order
Buyer confirms
STOCK → REFILL ↗
Predictive refill
Refill before stockout
Modules exist at different maturity levels. Automatic outbound proposals and unattended stock replenishment are integration goals—not a fully launched end-to-end service.
04 / BUYER-SPECIFIC QUALITY
One commodity. Different definitions of quality.
The valuable question is not merely “What is its score?” but “Is it right for this buyer and use case?”
See the EYEBOX prototype ↗
Soup stock
Extract · dryness · size
Prepared dishes
Color · mix · texture
Gift selection
Uniformity · shape · freshness
05 / REAL PRODUCTS, REAL OPERATIONSThe building blocks are already
The building blocks are already
in the field.
Imaging and matching prototypes, transaction and inventory software, and a real ordering channel exist. Next: integration, validation, and expansion.
01 / PRODUCT ECOSYSTEM
EYEBOX
Visual quality & buyer matchingImage analysis and buyer-matching workflow prototype
Explore
QuickInventory
02 / PRODUCT ECOSYSTEM
QuickInventory
Orders, inventory & operationsReal Android app connecting Kakao chatbot orders with sales, purchases and stock
Explore
03 / PRODUCT ECOSYSTEM
DasiWoorida
Real B2B ordering channelOnline retail/wholesale channel serving actual trade and order intake
Explore03 / QUICKINVENTORY · OPERATIONS PLATFORMOperated by MYSHOPCLOUD
Real order and inventory tools
for field operations.
From KakaoTalk ordering scenarios to web and mobile transaction records. The operational backbone in our Jeongsongi-centered product strategy.
Explore actual QuickInventory screens ↗
05 / THE REAL IR REPORT
Beyond analytics: a buyer-ready comparison.
An original comparison report from our company’s IR V3, showing price and visual quality criteria side by side.
Illustrative IR scenarioKRW 5,000 saved / boxField expertise + transaction history + buyer-specific preferences
MYSHOPCLOUD IR V3 · ORIGINAL REPORT · P.16View original IR report ↗
BEFORE
VSPROPOSED
Illustrative IR scenarioKRW 5,000 saved / box
Financial and conversion figures are illustrative scenarios from the company IR, not actual quotes.
07 / GLOBAL RAW MATERIAL STANDARDQuality standards must travel
Quality standards must travel
across borders.
Clothing sizes and shoe sizes enabled remote buying. Raw materials need mutually understood descriptions as well. We aim to replace unnecessary market visits with reliable images, traceable data and supporting documentation.
Food safety, microbiological and chemical properties require separate evidence and testing. Images are not a substitute for certified inspection.GLOBAL
QUALITY
LANGUAGE
QUALITY
LANGUAGE
08 / BUILD WITH USDemand and real customers exist.
Demand and real customers exist.
Now we need to scale what works.
We have field data, real customers, and working prototypes. We seek compute infrastructure, expert labeling, pilot partners, and capital to validate the next stage.
Explore partnershipsFROM REAL OPERATIONS
AI
TO A GLOBAL STANDARDFIELD EXPERTISE / CORE AI R&D / B2B DEMAND / GLOBAL POTENTIAL