Demonstrations & Technology Samples

An honest demonstration starts by being clear about what is being shown. We open every session by stating the model in use, the data it was trained on, the kinds of inputs we expect it to handle well and the kinds we expect it to mishandle. That framing earns us the right to ask stakeholders to bring their own examples — including the awkward ones — and prevents the demo from becoming a marketing event.

Where possible we run the demo on data the audience knows. A defect-detection demo on parts the quality team has personally inspected; a sentiment demo on a recent batch of contact-centre transcripts; a forecasting demo on the past quarter's actuals. The reactions in the room change immediately when the data is theirs, and the questions become useful in a way that they never are when the data is generic.

Sessions are deliberately structured for two-way conversation. We are interested in the questions the audience asks as much as in the model's responses to the inputs they bring; the questions are usually a more reliable guide to whether the technology is the right answer to their problem than any benchmark could be.

• Live, interactive sessions — not pre-recorded reels
• Audience-supplied inputs encouraged
• Clear statement of model strengths and weaknesses
• Anonymised use of audience data on request
• Follow-up notes capturing observations and limits
• Optional follow-on focused proof of value

Our demo library is updated continuously and currently includes vision demonstrations covering manufacturing defect inspection, document understanding (forms, invoices, identity documents, contracts), retail shelf analytics and visual similarity search; NLP demonstrations covering aspect-based sentiment, intent and topic classification, multilingual classification, retrieval-augmented question answering and named-entity extraction; forecasting demonstrations covering hierarchical demand forecasting, intermittent series, anomaly detection on operational metrics and probabilistic forecasting for inventory policy.

We also maintain a growing set of generative-AI demonstrations: grounded question-answering against curated knowledge sources, structured-output extraction from unstructured text, summarisation tuned for specific document classes, and assistant patterns scoped to particular workflows. We are deliberately conservative about generative use cases that would require open-ended generation in production; the demo library reflects the use cases we have actually been able to deploy responsibly.

Knowledge-systems demos round out the library: a worked rule-based decision system, a knowledge graph built from open data, and a hybrid neuro-symbolic example that pairs entity extraction with rules-based reasoning. These demos are particularly useful for audiences who have been told that machine learning is the answer to a problem that almost certainly is not best solved with statistics alone.

“The most useful demos are the ones that surface the question nobody had thought to ask.”

Where the standard library does not match the question well enough, we build a bespoke sample on your data within two to three weeks. This is not a proof of value; it is a sharper demonstration that uses your inputs against a pre-built model close enough to the use case to be informative. The output is a working prototype, a written assessment of what we observed, and a recommendation on whether a full proof of value is justified.

Custom samples are deliberately kept small in scope. We do not adapt models for production performance, we do not engineer the surrounding pipeline, and we do not build the user interface to a deployable standard. The objective is faster, cheaper insight than a full proof would provide; the trade-off is that the sample is not deployable as-is.

Many of our most successful client relationships started with a custom sample that surfaced an issue the original specification had not anticipated, and that re-shaped the subsequent project for the better.

For technical audiences we offer half-day or full-day lab sessions in which our engineers work alongside yours on a focused problem of your choosing. These sessions are part demonstration, part workshop, part candid conversation about what is achievable. They are particularly valuable for teams who are evaluating whether to build an internal capability or to engage external help; the conversation that emerges is usually more useful to the eventual decision than any vendor pitch could be.

Lab sessions are run under NDA where appropriate, and the artefacts produced — code, notebooks, evaluation results — are left with your team. We will not retain proprietary data shared with us in these sessions beyond the duration of the work.

Demonstrations are delivered remotely as a default, with on-site sessions available where the audience or the data warrants it. A standard remote demo runs sixty to ninety minutes and accommodates up to a dozen participants in a working conversation; larger sessions are possible but become less interactive. We are happy to deliver the same demo to multiple stakeholder groups separately if that helps internal alignment.

There is no charge for the standard demo library or for an introductory lab session; bespoke samples and extended workshops are quoted on the basis of the work involved. A short discovery call usually establishes the right format and content; we will not propose a session whose value is not clear to both sides.