Skip to content

Research

Applied Research & Experimental Development

Structured investigation of software, AI and computer-vision problems where the technical answer is not known in advance.

Research expertise

Where KRATT investigates

  • R1

    Artificial Intelligence & Machine Learning

    Dataset design, model selection and training, systematic evaluation and the integration of learned components into working software.

  • R2

    Computer Vision & Image Processing

    Detection, segmentation and geometric analysis of real-world imagery, including uncontrolled lighting, viewpoint and scene conditions.

  • R3

    Software Engineering

    Experimental and prototype software engineered well enough to produce trustworthy results, and to become the basis of production systems.

  • R4

    Measurement Systems & Technical Validation

    Deriving physical measurements from sensor and image data, quantifying error and validating results against defined acceptance criteria.

Definition

What KRATT considers a research engagement

Not every AI or software project is research. An engagement is treated as research only when the technical outcome cannot be determined in advance and the work is run as a systematic investigation.

Routine implementation, deployment and support are scoped separately from experimental research activities.

Technical objective
What the work must achieve, stated in technical terms that can be tested.
Existing knowledge reviewed
Published methods, available products and prior work are examined before anything is built.
Genuine technical uncertainty identified
The specific question a competent professional could not answer in advance from existing knowledge.
Candidate approaches formed
Hypotheses about how the objective might be achieved, and how each will be tested.
Experiments and prototypes run
Planned, versioned experiments rather than open-ended tinkering.
Results evaluated against defined criteria
Outcomes are measured against criteria agreed before the experiment.
Findings documented
What worked, what did not and why - including negative results.

Research process

Seven stages, recorded as they happen.

  1. 01

    Define the objective

    Agree the technical goal, constraints and the evidence that would count as success.

  2. 02

    Review existing knowledge

    Establish the current state of the art and what can be achieved with known methods.

  3. 03

    Identify technical uncertainty

    Isolate what cannot be known without investigation, and why.

  4. 04

    Form candidate approaches

    Develop testable hypotheses and an experiment plan for each.

  5. 05

    Experiment

    Build prototypes and run experiments, recording versions, data and configuration.

  6. 06

    Evaluate

    Measure results against the agreed criteria and decide whether to iterate, change approach or stop.

  7. 07

    Document findings

    Report conclusions, evidence and recommended next steps, whatever the outcome.

  8. Stages 04-06 typically repeat until the uncertainty is resolved or the approach is ruled out.

Availability

Research for independent organisations

Kratt Ventures undertakes commercially scoped research and experimental-development engagements for independent organisations. Engagements are priced on normal commercial terms and scoped according to the technical problem, expected investigation and required documentation.

Records

Research recordkeeping

Research engagements maintain project-level technical records covering objectives, uncertainty, experiments, versions, results, conclusions and relevant project activity.

Status

KRATT provides commercial research and experimental-development services in New Zealand.

Relevant experience

Selected prior experience of KRATT founder Martin Kambla

Computer vision for image-based timber measurement

Years of investigation into detecting and measuring logs from ordinary photographs: rule-based methods, learned detection, measurement pipelines and technical validation.

View case study

Have a technical question nobody can answer yet?

Describe the objective, what has been tried and any accuracy or performance target. We will tell you whether it is a research problem.

Discuss a research problem