Static walkthroughs, real artifacts

See how regulated AI actually ships.

These are static walkthroughs of artifacts Impetora delivers to clients. A contract clause that gets parsed into obligations. An evaluation harness output. A support conversation with the audit trail intact. Not live SaaS demos and not marketing renders.

For a live conversation about your specific problem, book a discovery call. We will prepare a recorded reference on a sample of your data before the call.

Reference showcases

Static

Walkthroughs of shipped artifacts

EU-hosted infrastructure

TRACE

Built using our methodology

Showcase 01

Document obligation extraction

A contract clause goes in. Obligations, risks, and quantitative thresholds come out, each with a citation back to the source line. Below is a static snapshot from a sample supplier agreement.

Input clause

9.3 The Supplier shall deliver the Services in accordance with Schedule 2 and shall meet each Service Level set out in Annex A. Failure to meet a Critical Service Level for two consecutive months gives the Customer the right to terminate without penalty on 30 days' written notice. The Supplier shall maintain professional liability insurance of not less than EUR 5,000,000 per claim.

Extracted

ObligationDeliver services per Schedule 2 and Annex A SLAs

ObligationMaintain professional liability insurance (>= EUR 5M per claim)

RiskTermination right after 2 consecutive Critical SLA breaches

Notice period30 days written notice for termination

QuantitativeEUR 5,000,000 minimum coverage

Static snapshot of a recorded run. Production deployments cite back to clause line numbers in the source file.

Showcase 02

TRACE evaluation harness

Anonymised snapshot from an evaluation harness output. Each test case ties a regulatory question to an expected answer, the model response, a pass or fail badge, and the citation chain. This is the artifact we ship before going to production.

Eval suite: regulatory-qa-v3

5 / 6 passing

EVAL-001Pass

Under MiFID II Article 25(2), when must a firm assess suitability for a retail client?

Expected: Before recommending a financial instrument or providing portfolio management.

Response: Suitability must be assessed before any personal recommendation or discretionary portfolio service is provided to a retail client.

Citation: MiFID II Dir 2014/65/EU, Art 25(2)

EVAL-002Pass

What is the maximum administrative fine under GDPR Article 83(5)?

Expected: EUR 20,000,000 or 4% of total worldwide annual turnover, whichever is higher.

Response: Up to 20 million euros, or 4% of the total worldwide annual turnover of the preceding financial year, whichever is higher.

Citation: GDPR Reg 2016/679, Art 83(5)

EVAL-003Pass

Does the EU AI Act classify a CV-screening tool as high-risk?

Expected: Yes, when used for recruitment or employment decisions (Annex III, point 4).

Response: Yes. Annex III, point 4 lists AI systems used for recruitment, candidate filtering, or employment-related evaluations as high-risk.

Citation: EU AI Act Reg 2024/1689, Annex III(4)

EVAL-004Fail

What is the cooling-off period for distance contracts under EU Directive 2011/83?

Expected: 14 calendar days from receipt of goods or conclusion of service contract.

Response: Consumers have a 30-day right of withdrawal from the date of contract conclusion.

Citation: Source not cited; expected Dir 2011/83/EU Art 9(1)

EVAL-005Pass

Under DORA, who is in scope for ICT third-party risk obligations?

Expected: Financial entities listed in Article 2 of Regulation 2022/2554.

Response: Banks, investment firms, insurance undertakings, payment institutions, and other financial entities defined in Article 2 of DORA.

Citation: DORA Reg 2022/2554, Art 2

EVAL-006Pass

How long must KYC records be retained under AMLD5?

Expected: At least 5 years after the end of the business relationship.

Response: A minimum of 5 years following termination of the business relationship, with possible extension to 10 years.

Citation: AMLD5 Dir 2018/843, Art 40

Static snapshot. Real evaluation suites are sized to the client problem and re-run on every model or prompt change before promotion.

Showcase 03

Customer support triage

A scripted exchange showing how an AI agent triages a refund request, looks up the order, applies policy, and resolves without human escalation.

Hi, I bought a coat last week and the lining is coming apart at the seam. Can I return it?

Sorry to hear that. Items with a manufacturing defect are eligible for a full refund or replacement within 60 days of purchase. Could you share your order number so I can pull up the case?

Order is #A-44218

I see it: coat ordered 6 days ago, EUR 189.00. I am opening a defect return. You will receive a prepaid label by email in two minutes. Refund is processed when the warehouse confirms the defect, typically within 3 working days of arrival. Anything else I can do?

No that's perfect, thanks.

Thank you. Have a good day.

Resolved in eval set

74%

Median time

42s

CSAT in eval set

4.6 / 5

Scripted reference flow on a fixed eval set. Production agents are tuned to your policy and tone.

Pipeline 01

Document extraction pipeline.

Every extraction passes through four stages with the audit log capturing inputs, outputs, model version, and reviewer action at each step.

Document obligation extraction. Every stage writes to the audit log.

Pipeline 02

Evaluation harness pipeline.

The harness takes a frozen test set, runs it against the candidate system, scores responses against expected answers and citation requirements, and gates promotion. Nothing reaches production unless the eval set passes.

Evaluation harness. Each run is logged with model version, prompt hash, and per-case scores.

Pipeline 03

Support triage pipeline.

The agent classifies intent, looks up the order, applies policy, and either resolves or escalates with the full conversation handed to a human reviewer.

Support triage. Resolution and escalation paths both write to the audit log.

Every showcase on this page was built using TRACE. Trust, Readiness, Architecture, Citations and Evidence are visible in every artifact, not bolted on.

Impetora delivery principle

Read the methodology behind these showcases at /methodology.

Demo FAQ

Frequently asked questions

Are these real Impetora systems?

Each showcase is a static walkthrough of an artifact we ship to clients: a contract extraction snapshot, an evaluation harness output, and a scripted support flow. The data shown is sample data, not client data. Production deployments run on the customer's documents under their data residency and access controls.

Why static walkthroughs and not a live demo?

Live demos hide the audit trail. We prefer static snapshots with the input, the output, the citation chain, and the underlying pipeline visible in one frame, because that is how we ship to clients. The TRACE methodology is the demo, not a chat widget.

Can I get a custom walkthrough on my data?

Yes. Book a discovery call with the workflow you have in mind. We typically prepare a recorded reference on a sample of your data before the call so the conversation is anchored in something you can see.

Can I share these showcases internally?

Yes. The page is public. Link colleagues to /demo or to the specific section anchors. Each showcase has its own pipeline diagram you can reference in your internal write-up.

Do these systems work in languages other than English?

The document extraction and support triage examples have been deployed in English and Lithuanian. Other languages are in scope on a per-engagement basis. Book a discovery call with the language requirements and we will scope the eval set accordingly.

Want a custom walkthrough on your data?

Book a discovery call and we will prepare a recorded reference on a sample of your data before we speak. Email info@ainora.lt or use the form on the homepage.

Book a discovery call