Services

Get from a difficult question to a useful next move.

We take on focused AI product and language technology problems: deciding what to build, finding what breaks, and making the result work in a real workflow.

Product & strategy

Start with the decision.

Bring a fuzzy opportunity, a stalled prototype, or a product that has outgrown its first assumptions.

Focused engagement

AI feasibility investigation

For a team asking whether AI belongs in a workflow at all, and what the smallest credible test would be.

  • Clarify users, source material, constraints, and failure costs.
  • Compare plausible model, rules, and human options.
  • Define a bounded pilot and the evidence it should produce.

Typical shape: 8–15 hours; a decision memo and test plan.

Discuss feasibility
Focused engagement

System audit & action plan

For an existing assistant, classifier, search system, or content workflow that looks promising but behaves inconsistently.

  • Trace the user journey and system boundaries.
  • Inspect a representative sample of failures and operational friction.
  • Prioritise fixes by impact, effort, and risk.

Typical shape: 12–20 hours; a ranked action plan.

Bring a system
Language & quality

Know what “works” means.

Aggregate scores hide the cases that matter. We make quality inspectable at the level of users, tasks, and failure modes.

Model and workflow evaluation

Build a practical evaluation scheme for language systems: test sets, error categories, human review criteria, and repeatable comparisons.

  • Define success, abstention, and unacceptable failure.
  • Sample difficult cases across languages, users, and contexts.
  • Compare approaches and turn findings into product decisions.

Typical shape: 15–30 hours; evaluation assets and a decision readout.

When the model is only half the product

We look at retrieval, source traceability, interface, handoff, and feedback loops alongside model behaviour. The result should help the team improve the whole system.

Especially relevant to knowledge assistants, community intelligence, editorial tools, and education products.

See the approach
Ongoing support

A product lead who can read the model.

For a team that needs technical product judgment without hiring a full-time specialist yet.

Fractional technical product leadership

Shape the roadmap, evaluation priorities, data contracts, and decisions between engineering, product, and domain experts.

Typical shape: 6–12 hours per month, with an explicit remit and decision cadence.

How delivery works

Signal & Sense is a founder-led specialist practice. Kate takes on a limited number of engagements. When a build needs more capacity, implementation can be separately scoped with named collaborators or partners.

You will know the scope, who is responsible, and what is being handed over.

Research & education

Make the knowledge usable.

Research and training can be the right intervention when the main obstacle is understanding, not another model endpoint.

Applied research

Frame a question, choose a defensible method, analyse real material, and translate the result into a decision or product opportunity.

Relevant to language use, information integrity, user needs, and AI system behaviour.

Courses and workshops

Design learning around real tasks: evaluating AI outputs, working with sources, understanding language systems, or building responsible editorial and research workflows.

We tailor the format to the audience and the decision they should be able to make afterwards.

Media and education are particular strengths. The same methods also travel to healthcare, community intelligence, legal information, and creative tools when the problem fits.

Next step

Bring the problem.
We’ll find the scope.

A short note about the users, current workflow, and sticking point is enough to begin.