Reconstructed from real BTI dataReconstructed from real BTI data — The store scenes on this page are reconstructed by our own visualization system from real BTI measurements.

Hero video · coming soon

매장 스캔 · BTI · 생성형 AI 데모

The real picture inside the store, in three seconds.

On-device AI · real-time retail vibrancy

One phone sweep — interior density and the queue at the door, measured. The scene reconstructed as a generative AI image. All recognition stays on the device; video never leaves. Built in Korea, ready for retail anywhere — designed for founders and property owners, not only trained analysts.

For investors For landlords

AI-generated · BTI grid sample

41 customers · 115 m² · BTI 12

scroll

There is enough data. There are not enough readers.

Legacy retail analytics

Analyst dashboard · not intuitive

RealDataLab

Generative AI · BTI 8 · 59 m²

AI scene · readable at a glance

Same store, same hour — shown two ways.

Five-year survival for Korean small businesses is 40 percent — five points below the OECD average. Roughly a quarter of closures begin with a wrong location call. Plenty of retail analytics already exist, but they are built for trained analysts. Aspiring founders, franchise managers, and property owners rarely make sense of the charts in time.

40%
한국 소상공인 5년 생존율
중소벤처기업부 · OECD 평균 대비 -5%p · 2024
53.6%
한국 소상공인 3년 생존율
중소벤처기업부 · 2024
29.7%
소매업 폐업률
중소벤처기업부 실태조사 · 2024
15.2%
음식점업 폐업률
중소벤처기업부 실태조사 · 2024

See where legacy tools fall short

Three firsts, delivered together.

There are three gaps even the global leader (Placer.ai) does not close: visualization for non-experts, real-time interior density per square meter, and direct waiting counts. We solve all three with on-device computer vision and patented generative AI.

Pillar 01

Visual intuition

Data as a photograph

We replace numbers and charts with an AI image of the store. Readable at a glance, even without analyst training.

Pillar 02

Bustle Index (BTI)

How busy is this store right now

BTI counts customers per 10 py (about 33 m²) of floor area. Not historical foot traffic estimated from GPS — the interior, right now.

Pillar 03

Measured waiting

A direct count of the queue

We measure the queue in front of the store — a first for the industry. The strongest signal of customer-experience quality.

See it in motion

Ninety seconds of the product at work.

One video walks through the full path — a single store scan, BTI computation, and a generative AI reconstruction. For visitors arriving via business-card QR, the demo is the second moment: visual proof that the system already runs.

Demo video · coming soon

Store scan · BTI · generative AI in one walkthrough

00:00-00:20
Store scan3-5 s sweep · 90 frames captured
00:20-00:45
On-device inferenceDetector + Tracker · NPU delegate
00:45-01:10
Four-step quality gateCoverage · Motion · Scan · Detection
01:10-01:30
BTI + AI imageCustomers per 10 py + reconstructed scene

How it works

After the three-second scan.

1
Scan
3-5 s phone sweep
2
On-device detection
Detector · NPU delegate
3
On-device tracking
Tracker · NPU delegate
4
Quality gate (4 steps)
Coverage · Motion · Scan · Detection
5
BTI + generative image
Customers per 10 py (≈33 m²)

All recognition runs inside the phone. Video does not leave the device — only counts and metadata. Detector and Tracker both run on LiteRT through the NPU delegate, reaching 30 FPS real-time inference on a Galaxy S22 and above. Only data that clears the four-step quality gate enters the dataset, and that data passes through a patented mapping algorithm into a reconstructed image of the store.

device telemetry

> scan started
> frames captured ........ 90
> detector ................ NPU delegate
> tracker ................. NPU delegate
> inference per frame ..... ~33ms
> quality gate ............ 88 / 100
> persons (unique) ........ 33
> store area .............. 99㎡ (≈30 py)
> bustle_index ............ 11.0
> upload (meta only) ...... 142B
> video ................... 0 byte

on-device inference · NPU delegate · video uploaded: 0 byte

Technical whitepaper

What good data looks like

What makes data good.

Data does not fail because there is too little of it. It fails when it does not reach the person making the decision. We define good data by three criteria.

It is easy to understand.

Data that only trained analysts can read is, for the decision-maker, the same as no data at all. It should meet aspiring founders, franchise managers, and property owners in their own language.

It is intuitive and visualizable.

A number that stays in the head and never becomes a picture of the actual store is rarely acted on. Decisions follow when the data leaves something that feels like a memory of having seen the place.

It carries weight in the decisions that matter.

Good data is not background reading. It is the deciding signal for site selection, property acquisition, and franchise-candidate choice. BTI numbers and AI store images together put data at the center of the decision.

Why a single number is not enough.

Human decisions are not driven by cold numbers alone. Antonio Damasio reported a patient (pseudonym Elliot) whose intact logical reasoning still left him unable to make ordinary daily choices after damage to emotion-processing regions of the brain. [c22] The Iowa Gambling Task showed that people with ventromedial prefrontal damage stop generating bodily signals (skin-conductance responses) before risky choices. [c23] Kahneman summarized decades of work showing that most everyday decisions run on fast, intuitive processing. [c24] The amygdala-to-vmPFC circuitry assigns value to incoming information ahead of conscious deliberation. [c25]

Images become the emotional anchor of a decision.

People recognize images they have seen before with over 90 percent accuracy. [c26] Across studies dating to the 1970s, pictures are processed faster than words and remembered longer. So we reconstruct the store as an AI image alongside the BTI number. The number makes places comparable; the image gives the decision an anchor.

Generative AI · BTI 12 · 115 m²

41 customers · 115 m² · BTI 12

Good data is easy to understand, visible, and central to the choice. RealDataLab is designed on those three lines.

Citations

[c22]Antonio Damasio · "Descartes' Error: Emotion, Reason, and the Human Brain" · G.P. Putnam's Sons · vmPFC 손상 환자 Elliot 사례
[c23]Bechara, A., Damasio, A. R., Damasio, H., & Anderson, S. W. · "Insensitivity to future consequences following damage to human prefrontal cortex" · Journal of Neuroscience 14(5): 2727–2740
[c24]Daniel Kahneman · "Thinking, Fast and Slow" · Farrar, Straus & Giroux · System 1/2 (Stanovich & West 프레임 대중화)
[c25]Joseph LeDoux · "The Emotional Brain: The Mysterious Underpinnings of Emotional Life" · Simon & Schuster · 편도체 빠른 정서 처리
[c26]Lionel Standing · "Learning 10,000 pictures" · Quarterly Journal of Experimental Psychology 25: 207–222 · 이미지 재인식 90%+ 정확도

Where existing data falls short

What retail data has been missing.

The market is not short of tools — the global leader carries a $1.5B valuation, and several Korean tools have been around for years. Yet small-business closure rates do not improve. Three gaps coexist.

Location precision — the inside of the store is invisible.

Legacy

GPS and Wi-Fi estimation works at roughly 50 m. A first-floor cafe and a fifth-floor office in the same building look the same. After iOS 14+ and Android 10+ randomized MAC addresses, Wi-Fi probe tracking has lost its foundation.

RealDataLab

Camera-based interior measurement. Different stores at the same address are separated at pixel-level resolution per square meter.

Spending intensity — credit-card data misses about 37 percent.

Legacy

Per the Bank of Korea Payment Methods Usage Survey 2024 (released 2025-03-25), the offline payment mix by transaction count is credit card 46.2%, debit card 16.4%, mobile card 12.9%, cash 15.9%, and other 8.6%. Analytics built on credit-card sales miss the remaining ~37% (cash, digital wallets, and other), while the 5–7 day card-settlement cycle adds a real-time gap and digital-wallet adoption among younger cohorts introduces a generational sampling bias.

RealDataLab

We count interior occupants directly instead of inferring from payments. BTI is customers per 10 py (≈33 m²) — a direct measurement independent of payment method, settlement cycle, and generational distribution.

Queue signal — nobody measures it.

Legacy

From Placer.ai down to local Korean tools, no commercial service measures the queue in front of a store. The strongest signal of "a store people line up for" never reaches the data.

RealDataLab

Field surveyors count the queue directly and the count enters the dataset. The most powerful signal of customer-experience quality, quantified for the first time at this scale.

A second limit specific to Korean tools.

Nice Bizmap, SK Geovision, and Ministry-of-Public-Administration open data are all post-hoc aggregations. Time-series lag runs six months or more, store interiors are invisible, and visualization stops at analyst dashboards. A founder choosing between leases next week cannot lean on six-month-old aggregates.

See the full table + four structural limits

Source: proposal AX 2026 §3-2 · 재도전성공패키지 business plan · Bank of Korea payment statistics

Who uses it.

Bring data to a vacant lot. A three-second scan instead of a four-week consulting engagement. Vacancy in Myeongdong sits at 4.4 percent, Hongdae at 10 percent — we put micro-location value back on the negotiation table. Video stays on the device; only metadata travels.

Property diagnostics

Case studies in preparation

미국 Location Intelligence 시장 · 2025

$25.4B

Grand View Research

60%+

Reanin Research

Location intelligence · global · Reanin / GVR

Why now

The market is ready.

$8.65B

미국 Foot Traffic Intelligence 시장

Reanin Research · Wi-Fi 분석 등 포함 · 2025

$25.4B

미국 Location Intelligence 시장

Grand View Research · 14.7% CAGR (2025~2030) · 2025

310조원

한국 프랜차이즈 시장

한국공정거래조정원 · 한국프랜차이즈산업협회 · 2025

14.7%

미국 Location Intelligence CAGR

Grand View Research · 2025-2030

100만+

한국 자영업 폐업 (2024)

중소벤처기업부 · 통계 집계 시작 이래 최초 100만 돌파 · 2024

25%

폐업 원인 1위 — 입지 선정 실패

소상공인시장진흥공단 · 언론 설문 · 2025

Location data is already a large industry in the United States. Foot-traffic intelligence alone is $8.65B; the broader location-intelligence market sits at $25.4B and is growing 14.7 percent a year. In Korea, franchise transactions move ₩310 trillion annually, and 1.13 million people start a business each year. Yet no commercial service anywhere measures real-time density inside the store. That is where we begin.

Sources: Reanin Research · Grand View Research · Korea Franchise Council · Ministry of SMEs · Small Enterprise Promotion Agency

Defensibility

What makes us hard to copy.

Our moat is not a single line of code. It is the patent family, the data network effect from crowdsourced surveyors, and the sequence — prove it in Korea, then take it to the United States. More data sharpens the generative model; a sharper model brings surveyors in faster.

Patent #01KIPO filed

Sensor fusion + four-step quality gate

Measuring the trustworthiness of the survey itself

Gyroscope, accelerometer, and ambient-light fusion automatically scores survey quality. Only scans above 80 enter the dataset, so crowd-sourced data quality holds.

Patent #02KIPO filed

BTI → visual parameters · 3-trigger generative synthesis

A global first

Area, category, and time-of-day metadata become visual parameters; three triggers (store / building / district) isolate the generative AI synthesis. The core technique that makes the store legible at a glance.

* Both patents are in active KIPO examination. The page updates on registration.

Progress

Already running.

Phase 1 end-to-end was running in March 2026. Phase 2-A is in progress as of May — AI pipeline, surveyor system, generative imaging, and the customer web app are converging. Every line was written by a single founder paired with Claude Code; the GitHub history is the receipt.

Roadmap

Phase 1
✅ 2026-03
E2E running
Phase 2-A
In progress
AI pipeline + customer web
Phase 2-B
Next
Camera-stack transition
Series A
Planned
Capital ramp
NASDAQ
Long term
Global Select Market

Founder

Who builds it — and who stands behind it.

Fourteen years in large enterprise. Father of three. Full-time founder since 2026. Most of Phase 1 and 2 was implemented directly. Not originally a developer — pairs with AI to write code. That is the signature here: one person responsible end to end.

Request a meeting by email

realdatalab@realdatalab.co.kr

Verified facts

Phase 1 end-to-end

Running March 2026

Camera input through BTI computation and server transmission, in one verified flow.

Two KIPO patents

Filed

Sensor-fusion quality gate + BTI-driven generative AI visualization.

AI pipeline

On-device, verified

Detector and Tracker running in real time on the smartphone NPU.

Full-stack, solo

WSL · Android · Web · Backend · DB

A one-person full-stack pipeline assembled through Claude Code collaboration.

Korean retail data, recognized in the United States.

"Build an app that is loved by many. Make the structure difficult for competitors to imitate. List on the Nasdaq Global Select Market — the market that values location intelligence most."

— RealDataLab, Vision §1.2

Watch the demo Request a diagnosis

NASDAQ Global Select Market