Product demos gallery
SimulationsTry the patterns behind the products
These are demonstrations, not live products. Each runs entirely in your browser on synthetic data, with no AI model and no data sent anywhere. They show the control logic we build around models: thresholds, citations, abstention and approval.
Simulation 01
Document extraction with validation
Pick a synthetic invoice, run extraction and see how confidence thresholds and validation rules decide what goes to review.
Values and confidence scores are pre-written synthetic data. The point is the routing logic, not the extraction.
Choose a document and run extraction.
Related product concept: Document Intelligence
Simulation 02
Cited retrieval over a policy library
Ask a question over a small synthetic policy library. A keyword ranker retrieves passages and the answer quotes them with citations, or abstains.
Keyword ranking (BM25-style) over eight invented policy passages. The answer is extractive: it quotes a retrieved passage, it does not generate text. It abstains when the best passage does not cover most of the question.
Answer
I could not find this in the policy library. Please ask the HR or policy owner. (Top retrieval score 3.71 is below the evidence threshold.)
Retrieved passages
- [1] Expenses policy §1.23.71Expense claims must be submitted within 30 days of the expense date with an itemised receipt. Claims older than 90 days are not reimbursed.
- [2] Expenses policy §4.12.71Client entertainment requires prior approval from a budget holder and must list all attendees on the claim.
- [3] Travel policy §2.10.00Economy class is required for flights under six hours. Premium economy may be booked for flights of six hours or more with manager approval.
Related product concept: Enterprise Knowledge Copilot
Simulation 03
Agent proposal and human approval
Step through an agent working on a synthetic support ticket. Read-only steps run automatically; the write step waits for your decision.
Synthetic ticket #EX-3141
“We can’t add our new analyst. The invite button is greyed out.”
Read-only steps run automatically. The single write step stops for a person.
Run trace
No steps yet.
Related product concept: AI Agent Workspace
Want to see the real thing on your data?
A simulation shows the pattern. A discovery sprint shows whether it works on your documents, questions or tickets.