The criteria live in documents
Policies, operating standards and contract terms sit in a document repository. Half of every answer is finding and reading that document.
Documents and data
in a single question.
Whatever must be masked stays masked wherever the query runs. When you outgrow it, PAASUP DIP takes over.
After the consultation we run a PoC on one data source.
What stalls is not the model.
It is governance.
Pilots mostly succeed. The moment you widen them, it stops at who may see what. Permissions, sensitive data and scattered sources arrive at once, and the industry is stalling in the same place.
Failures are most likely to occur when organizations fail to distinguish between an agent's ability to act and the scope of access it is granted. By 2027, 40% of enterprises will demote or decommission autonomous AI agents due to governance gaps.
By 2027, 60% of organizations that ignore data governance culture challenges will fail to govern AI successfully.
Writing a policy and making that policy actually apply on every read path are two different jobs. PAASUP DNA takes the second one.
One system
If those three live in three different tools, each path returns a different answer.
01
Run AI inside the permission boundary
People and AI agents pass the same gate. What something may do and what it may see are set separately.
02
Use sensitive data safely through masking
Rather than blocking it, let people use it masked. The policy applies at read time and the original is untouched.
03
Analyse documents and databases as one
Criteria live in documents, numbers live in the database. Analysis only becomes routine when the asker need not know which.
Documents here, databases there, masking somewhere else.
One question means three systems.
The numbers live in the database
How many, how much, right now: that is in the database. Getting the number means going through someone who writes SQL.
Masking is set up per system
When each system masks personal data its own way, the same person sees different values depending on the path they took.
A question that used to cross three systems ends in one.
Document repository, data catalog and de-identification policy live in one environment. The person asking does not need to know which system holds the answer.
Documents and the database answer together
Documents, unstructured and structured data live in one repository, one catalog, one question box. One question gets an answer from documents and the database together, with the source file name, page and the SQL that ran.
Whatever must be masked stays masked wherever the query runs
Personal data is masked at read time by a policy set once. Document queries and SQL results fall under the same policy. The original is never touched.
People and AI agents see only what they are allowed to
Viewing, search, SQL and ETL all pass through the same permission gate. Run the same query as two users with different permissions and different values get masked.
Shown on real screens.
Captured on a demo tenant. All personal data is fictitious.
Documents + data in one conversation
The criterion from the document, the current figure from the table. One answer carries the source file name, page and the SQL that ran.
Content filter bindings
doc:// document paths and data:// table columns share one de-identification policy. Exceptions only for named subjects. Anyone who is not an exempt subject sees masked values only.
Structured-analysis notebook (Korean UI)
Several SQL cells in one document, each rendered as a different chart (line, bar, pie). Switch chart types per cell and try them out.
MD dashboard (Korean UI)
The same data laid out as KPI, trend, comparison and share cards in one dashboard document. The document itself is the dashboard.
Large-scale processing is delegated to PAASUP DIP.
Batch, virtualization, streaming and LLM serving move to the platform below, and permissions and policy carry down as is. Who can see what does not change. See PAASUP DIP
Start with one data source.
-
1
Consultation
We pick one data source and one document repository.
-
2
PoC
Run the same question under two permission sets within that scope. That is where the PoC ends.
-
3
Grow
Add sources and repositories as you need them. Permissions and policy follow the same way.
Questions we get before a review
Anything that depends on your environment is answered in the consultation.
Which databases can you connect to?
Relational databases such as PostgreSQL and MySQL, S3-compatible object storage, and REST APIs. A PoC starts with a single source, so you do not have to finish every connection first.
Do you copy our source data?
Masking is applied at read time and the original is left untouched. Whether a separate analytical copy is made is decided per environment at rollout.
Is our data used to train external models?
No. For large-scale processing or air-gapped requirements, model serving can run inside your network on PAASUP DIP.
What happens when two people with different permissions ask the same question?
They get different values back. Viewing, search, SQL and ETL all pass the same policy gate, and only explicitly named subjects see unmasked values.
How do we check that an answer is trustworthy?
Every answer carries the source document name and page, plus the SQL that was actually executed. If a number looks wrong, open that SQL.
How long does a PoC take?
Pick one data source and one document repository in the consultation, then run the same question under two permission sets. That is the PoC.
Tell us which question
you want to ask in one go.
Share the questions you ask often and where their answers live in documents and data. We will reply with a scoped PoC.
- One or two questions that take too long to answer
- Where the related documents and databases are
- The kinds of information that must be masked (personal data and so on)