Home Benefits How It Works Platform Contact
How It Works

Define what a good answer needs.
Check the evidence.
Then - and only then - generate.

Most AI systems generate first and check later - if at all. enSmaller works the other way round. It determines what the answer should contain, filters the evidence, and only proceeds when there is enough to support it.

In simple terms

enSmaller does three things:

Decides what the answer needs to include
Only gives the model the evidence needed for that
Blocks or explains anything that can't be supported

The model sees less irrelevant content, so it produces more reliable answers. Cost drops because fewer tokens are processed. And every output is traceable - because the system already tracks what was required, what evidence was used, and what was verified.

The process

Four steps, every time.

Step 1

Decide what the answer needs to include

The system analyses the question and determines what a good answer must cover, what type of evidence is needed for each part, and what standards apply.

Step 2

Find and filter the evidence

Evidence is retrieved and filtered - not just for relevance, but for whether it comes from the right source and is appropriate for the specific part of the answer. Irrelevant or low-quality material is excluded before the model sees it.

Step 3

Check there is sufficient evidence before generating

Before the model is called, the system checks whether there is enough reliable evidence to support each part of the answer. If something is missing, it reports the gap - rather than letting the model fill it with guesswork.

Step 4

Verify the output

After generation, each part of the answer is checked against the evidence: is it supported, is it complete, and did it use appropriate sources? If something fails, that gap is surfaced explicitly.

The difference

What changes when enSmaller is in the pipeline

Without enSmaller

1 Retrieve content by similarity
2 Send everything to the model
3 Generate an answer
4 Check it with a single score
5 If something's wrong, you find out later

With enSmaller

1 Define what the answer needs to include
2 Send only what's needed - nothing else
3 Check evidence is sufficient before generating
4 Verify each part of the answer independently
5 If something's missing, tell you exactly what and why

The governance that makes outputs trustworthy is the same mechanism that reduces token usage. Less noise in means better answers out - and a lower AI bill.

What you get

Every output comes with:

Requirement-level verification

Not a single "grounded" score. Each part of the answer is assessed against explicit requirements - so you can see what is supported and what isn't.

Evidence chain

Which sources were used for which claims, and how they contributed to the answer - giving you a clear link between outputs and evidence.

Gap reporting

When information is missing, the system identifies what cannot be answered and reports the gap - instead of filling it with assumptions.

Decision log

A detailed record of what was required, what evidence was used, and how the answer was evaluated - suitable for audit, debugging, or review.

See it working on your data

Bring us one contained use case where output quality, compliance, or cost matters. Real data. Real requirements. We'll run a short paid pilot through enSmaller and show you exactly what changes - what gets caught, what gets refused, and what it saves you.

Get in touch