Vendor-neutral
One standard for every AI.
ChatGPT, Claude or any other model: the engine applies the same method and the same standard to every answer.
The zenture engine
Not another model. An engine built to judge.
zenture evaluates AI answers on an engine we designed and built ourselves: proprietary logic, its own data and dedicated infrastructure in Germany. It works independently of the model that wrote the answer and returns one decision you can act on.
See pricingWatch it work
One answer in. One decision out.
How one evaluation runs
Five steps from answer to decision.
01Understand
Your task becomes a contract.
A model reads your request and turns it into a precise task contract: what was asked, which requirements must hold and which standard applies.
02Take apart
The answer is split into signals.
The engine separates the answer into its claims, its cited sources and the requirements it has to meet, and plans the checks each one needs.
03Retrieve
Evidence is fetched under a strict policy.
Fixed rules retrieve sources: public HTTPS only, private and internal addresses refused. What cannot be reached is marked unavailable, never guessed.
04Weigh
Every claim meets its evidence.
Language models weigh each claim against the retrieved evidence. Every model output is validated against a strict schema before the engine accepts it.
05Decide
The verdict follows fixed rules.
Ambiguity goes to human review. Missing evidence means not enough evidence. Unmet requirements or findings mean revise. Only a clean pass is ready.
Four layers
One machine, four layers working as one.
01
Algorithms
zenture’s own algorithms form the verification logic. They retrieve sources, run checks and turn the findings into a verdict according to fixed rules.
02
Data & configurations
zenture’s own data and configurations define the checks and their context. Together with your task’s requirements, they give the algorithms a precise basis for each evaluation.
03
Model
An open-source language model, hosted, configured and deployed by zenture, helps interpret your task and weigh the evidence. The engine validates its outputs before applying the decision rules.
04
Infrastructure
zenture operates the infrastructure that brings algorithms, data, configurations and the model together as one engine. It runs each evaluation from input to verdict and stores your zenture data on servers in Germany.
The result
Four clear outcomes. Always a next step.
Ready
The answer holds up against its requirements and the evidence. Use it.
Revise
Requirements are unmet or findings need fixing. You see exactly what to change.
Human review
The case is ambiguous. A person should make the call, with the evidence laid out.
Not enough evidence
Sources are missing or unreachable, so the claims cannot be confirmed yet.
Every result lists its findings with their confidence, the evidence behind them, known limitations and the next step to take.
Independent by design
It judges answers. It does not write them.
Independent
Not the model grading itself.
zenture never asks the model that produced an answer to judge its own work. The evaluation runs on zenture’s own engine.
Your data
Hosted in Germany. Not used for training.
Dedicated zenture infrastructure stores your zenture data on servers in Germany. Customer conversations are not used to train a general-purpose AI model.