Screen every application. Defend every decision.
A panel of AI analysts reads each application against your own rules and quotes its evidence. Code checks every quote. A person decides.
Illustrative example
Run record
Application round · 24 applications
- Intake and validation
- Panel scoring — five members
- Human approval
- You choose where your data is processed
- Every step timestamped, every decision sealed
- Or run it on a dedicated computer
Built for organisations that have to justify their decisions
If you receive applications and must show why each one passed or failed, this is made for you.
Foundations and grant funders
Find the applications that break a rule before an assessor reads a page. Every result points to the passage it rests on.
Grant applicationsSchools and universities
Check scholarship, grant and course applications against your published criteria. The assessor gets a quoted summary and makes the decision. MndUP never ranks applicants for admission.
Scholarship applicationsConsultancies and agencies that review for clients
Run screening for many clients from one account. Each client keeps its own rules, results and records, apart from everyone else’s.
Use casesPublic agencies and municipalities
Pre-screen grant applications against the rules that apply, with data processed in the EU only when you choose that setting. Case officers get the facts and the quotes; the decision stays theirs.
Data and security
Anything you must check against written rules
Grant and application screening is proven. For everything else we run a pilot with you first.
- Course applications
A person decides.
- Your own documents
Question answering with citations over your own documents exists in the software; a packaged offer does not yet.
For anything marked pilot, we agree with you what to test and what a good result looks like before you commit.
From your rules to a sealed decision in four steps
Your rules go in. A decision you can defend comes out.
You write the rules
Criteria, scoring scale and screening rules are data in an editor you control. No developer, no template.
A panel of AI analysts reads
Several analysts read each document. Every score comes with a written reason. Every quote is checked against the source, and a quote that cannot be found is flagged.
Code checks every quote
Software finds each quote in the source. One it cannot find is marked unverified and stays in view.
A person decides
The assessor accepts, changes or rejects. The decision, the evidence and the time go into a sealed record.
Sort the pile before anyone reads it
Every application lands in one of three groups, each with its reason in plain words.
Stopped
The application breaks a rule. The rule and the reason are stated in plain words.
Check by hand
The software cannot settle it. A person looks.
Passes
No rule was broken. It moves on to full assessment.
Every stop shows its rule and either a checked quote from the application or a fact from your own data. Nothing is stopped without a reason you can read and challenge.
The record that answers “why was this rejected?”
One file holds the whole assessment: each analyst’s score, where they disagreed, who approved and when, and the reasoning in writing. Sealed with a hash and a signature. This is the file itself, not a dashboard about it.
Illustrative example — a synthetic application, shown in the real record's structure
Strengthening local democracy through youth councils
Reference GR-26-047 · Application round 2026 · 1–6 scale · Five AI panel members
Scores by criterion
Relevance to the call
Feasibility of the plan
Organisational capacity
Expected impact
Where the panel disagreed
Four panel members scored organisational capacity 4; the fifth scored 2, citing an unaudited financial statement for the previous year. A spread of 2 meets this round's threshold, so the application was flagged for a person and entered the review queue before the round closed.
Human approval
Approved by the round's assessor, who saw the divergence, the dissenting member's written reason, and the application itself alongside every score.
Written justification
The plan is concrete, names its partner schools and sets a realistic timeline, and its relevance to the call is direct. Organisational capacity is the weak point: last year's financial statement is unaudited and the projected staffing assumes a funding decision by June. Recommended for funding on condition that an audited statement is filed before any payment is made.
The exported jury sheet states, in the file itself, what the ranking is: assistive triage, dependable at finding the weak applications and unreliable at ordering the strongest.
The AI never has the last word
Every application leaves the panel by one of three routes. None of them is “the model decided”.
Agreed — and still sampled
Where the panel agrees and the application sits clear of the funding line, its draft stands. A fixed share of these is routed to a human anyway. That share defaults above zero, and the round records the rate it used.
Disagreement goes to a person
A wide spread between analysts, weak agreement, a criterion only one could score, or a rank near the funding line sends the application to a person. Where the round is set to pause, the run stops there until someone looks.
Below the line — and still sampled
Applications the panel places below the funding line carry the same oversight sample as the confident ones. Being ranked out is not a reason for nobody to look.
MndUP is designed for human oversight, and makes no claim of legal conformity. The oversight is built into the software and tested, not promised in a policy document: the share of confident cases sent to a person starts above zero.
What sets MndUP apart
Every line is built and running today, except the ones tagged coming.
Every quote is checked
Checks every quote against the source. A quote it cannot find is flagged, never hidden.
A person decides, and the decision is sealed
Seals every decision with a hash, an audit chain and a signature, exportable as JSON or PDF.
A panel of AI analysts, not one model
Convenes several AI analysts per document and demands a written reason for every score.
You choose where your data is processed
Enforces your data-processing choice on every model call.
Your rules, measured against your past decisions
Measures the analysts against your own past decisions, so you see how they compare with your assessors.
Evidence
Agreed cases still sampled
Sends a share of the cases the analysts agree on to a person too.
Every screening rule in one view
Lists every screening rule in one Criteria view.
Control
Your process, in your hands
Puts the whole process in your hands as editable data, no developer needed.
Cost shown before a run
Shows an estimate of what a run will cost, per document, before it starts.
Each client kept apart
Keeps each client's data apart for consultancies.
Running it
Runs on a dedicated computer
Runs on a dedicated computer with the local edition, where models on that computer read the documents.
Re-run only what changed
Runs again only what changed, and shows what will be asked again before you start.
An API for every action
Offers an API for every screen action.
Your data is processed where you say
Each organisation decides where its data may be processed, and MndUP enforces that choice on every model call.
Each organisation chooses where its data may be processed.
There are three settings: anywhere (the default), EU only, or only on a dedicated computer with the local edition. The setting is enforced on every model call. Web and public-register lookups, and connectors you turn on (webhooks, Drive, storage exports), send their requests to those outside services.
Multi-factor authentication is available for every account.
Roles decide who can see, run and approve.
A data processing agreement is available on request.
The privacy notice lists what is stored and your rights. Read the privacy notice.
The questions your procurement will ask
Straight answers, including what we do not claim.
- Is MndUP GDPR-compliant, and do you offer a DPA?
- We design for GDPR: EU hosting, a residency setting enforced on every model call, roles and a record of each step. A data processing agreement is available on request. Your organisation remains responsible for its own lawful basis for processing.
- What about the EU AI Act and human oversight?
- MndUP is designed for human oversight: a person makes every decision, sees the evidence, and can overrule the analysts. We make no claim of legal conformity; whether a use is high-risk is for you and your adviser to judge.
- How are quotes checked?
- Code searches for each quote in the source document. A quote it finds is marked verified. A quote it cannot find is marked unverified and kept visible, so you see it rather than lose it.
- What does a pilot involve?
- A pilot runs your own rules on a set of your past applications and compares the results with your own decisions. The scope, the set of applications and what counts as a good result are agreed with you first.
See it on your own applications
Bring your rules and a handful of real applications. We run them in front of you, show the record that comes out, and answer what your procurement will ask.