AI that understands you

A work-specialized neural network, verified by an accredited test

ez2AI works with two neural networks: one on your PC that remembers you, and one that reads the structure and relationships of company data.

Technical consultation Where your data lives

The neural network on your PC

An Agent neural network that adapts to you the more you work

It ties your evidence to each task and adapts to you over time. This neural network lives only on your PC.

Ties evidence to each task
Links memory, persona, reference work, email, attachments and design to the task; the AI reads them before answering.
Remembers on the spot
Decisions, rules, corrections, style and format agreed in conversation become memory right away.
Frequently used memory first
Recently used memory and memory from the same project are brought forward for the next task.

The neural network for company data

Finding relationships in databases with no design docs

Even without foreign keys or comments, we verify by actual values, not names.

01
Shortlist candidates
Pick only columns likely to link
02
Table roles
Classify master, transaction, history
03
Match columns
Align name, type, meaning
04
Check value overlap
See if actual values overlap
05
Determine relationship
1:N, N:M direction and cardinality
06
Re-verify
Re-score earlier judgments

Knowledge becomes Auto-structured & auto-linked knowledge, viewed in 3D in real time

Knowledge is automatically structured. It is then auto-linked by category, hierarchy, and semantics — and visible in 3D, in real time.

Internationally accredited testing

Connect scattered company knowledge with the ez2AI-Core neural network

Connecting means finding what links to what, and how.A nationally accredited laboratory (KOLAS) verified that ability first-hand.

Report no.: KST-26-131 · issued 4 Aug 2026
Laboratory: KOLAS-accredited testing laboratory
Standard: ISO/IEC TS 4213:2022 6.2.4
Method: On-site — the examiner ran and scored the test at our premises
Scope: Automatic relationship recovery, with the answer key withheld

100%

Recall

Not a single relationship that should have been found was missed.

98.53%

Precision

The share of proposed relationships that matched the answer key.

It surfaces relationships the answer key never contained

In this test ez2AI-Core missed nothing the answer key listed (100% recall), and on top of that it flagged one relationship the answer key did not contain. Under the test rules that single item counted as a false positive, which is why precision reads 98.53%.

Out in the field, though, there is no answer key. In a system 15 to 20 years old the foreign keys were never declared and the people who knew the relationships have long since left. What you need there is not the ability to confirm relationships someone wrote down — it is the ability to surface the ones nobody ever did.

Flagging something outside the answer key means the model did not memorise a fixed answer; it read the structure of the data itself. We consider that single item the most important part of this test, so we publish both figures rather than hiding one.

People decide. Relationships the model proposes come with their supporting evidence and are presented for review, and your team confirms them. In legacy analysis, surfacing more and reviewing is safer than missing something.

The six years of real data behind ez2AI-Core

ez2AI-Core is built on 74,332 reviewed data records accumulated while building enterprise systems over six years. Repeated validation in real work became the model's foundation. This is not data generated arbitrarily by AI. Expert developers organized and verified the data structures used in real system projects.

01

Analyze real work

Analyze systems and workflows across different industries.

Work structure mapped

02

Accumulate proven rules

Standardize reusable data structures and relationships.

Shared asset built

03

AI core optimization

Intensively trained on 74,332 rigorously refined gold metadata datasets.

Irreplicable ez2AI-Core neural network established

74,332 gold metadata datasets accumulated through six years of enterprise system know-how with more than 100 companies

ez2AI-Core learns only the dataset structure and labels of the metadata.

All sensitive enterprise data — customer information, work data, operational DBs and more — is completely excluded from AI neural network training and rigorously protected.

How ez2AI-Core Classifies and Connects Enterprise Data

See how ez2AI-Core uses a 3D graph to classify enterprise data in real time and connect relationships between data.

Keeping the AI from improvising

Lock in company knowledge before building, then re-score against the design after.

Lock in company knowledge first
Inject data structures, menus, APIs and naming rules at build time.
Re-score against the design
Review the output again against the design spec.
No evidence? "Needs review"
Asks a person instead of inventing values.

Four principles for connecting to live systems

Work alongside your existing ERP and MES — no replacement, no downtime.

Read only
Reads the production database in read-only mode.
Never delete
Writes require approval, nothing is deleted, and history is kept.
Never invent
Without evidence, items are flagged "Needs review".
Never stop
If an external connection drops, work continues via another path.

Where does your data live?

What lives where, for each deployment option, in a single table.

Swipe horizontally to compare deployment options.

	Individual	Business	On-premises	Air-gapped
Work execution	User PC	User PC	User PC + in-house servers	In-house
Work files, chats, code	User PC	User PC	User PC, in-house	In-house
AI calls	Your account, direct from PC	Business AI seat, direct from PC	Business AI seat or in-house model	In-house model only
Sent to external AI	Instructions, results	Agreed per customer	Agreed per customer, or in-house model	None
Uploaded materials (Works)	Korean cloud, isolated	Korean cloud, isolated	In-house servers	In-house
Recording, PDF conversion	Not kept after processing	Not kept after processing	In-house	In-house

Air-gapped deployments require an in-house inference model and GPUs, and are priced separately after reviewing requirements.

Talk to us about the right way to get started

Tell us about your current workflow and the capabilities you need. We’ll recommend the right product and implementation scope.

Request a consultation

Frequently Asked Questions

Check answers to common pre-adoption questions on the FAQ page by category.

Technology | ez2AI data relationships and AI workflows