Human AI Judgment
Work with me

Build

Anything can be made to work once

The distance between that and a system an organisation can rely on is where the work is: the architecture underneath it, the data it depends on, the verification that keeps it honest, and the people who have to operate it afterwards.

An applied-AI system from unstructured inputs through data, models and verification to a result a person operates, and the loop back into the work.

01Featured

Building applied AI systems

I work across the path from an idea or operational problem to a system people can actually use: product architecture, data, machine learning, AI agents, computer vision and the capability needed to operate them responsibly.

  1. Problem
  2. Data
  3. Models
  4. Systems
  5. People
  6. Improvement

02Build areas

Five areas of work

Capability rather than tools. Each area is a chain: what goes in, what has to be decided along the way, and what comes out the other end.

  • 01

    Product & app development

    Product development from interface concepts through application architecture and data services to a person using the finished app.
    1. Product
    2. Architecture
    3. Data
    4. Interface
    5. User

    Applications taken from product logic through architecture, data and interface to something a person can actually use.

    The product decisions, the data model and the application architecture stay legible and under my own control, which is what makes the thing changeable after it ships.

  • 02

    AI agents & agentic systems

    An agentic system assembling context from mixed inputs, reasoning, reaching for tools, and returning a result an operator verifies.
    1. Input
    2. Context
    3. Reasoning
    4. Tools
    5. Output
    6. Verification

    Systems that can reason through a task, use tools and context, interact with a person and return something useful within defined boundaries.

    Boundaries first: what the agent may reach, what it has to ask, what a person still decides, and how the result is checked before anyone acts on it.

  • 03

    Machine learning & data

    A machine-learning workflow from raw data through representation and modelling to candidates that are evaluated before one decision is taken.
    1. Data
    2. Representation
    3. Model
    4. Evaluation
    5. Decision

    Turning data into models, and models into evidence that can support a decision.

    Framing the question, building the representation, training the model, and the part that decides whether any of it was worth doing: evaluating it honestly.

  • 04

    Computer vision & industrial AI

    An industrial computer-vision line from image acquisition through processing and a model to an inspected accept-or-reject decision.
    1. Image
    2. Processing
    3. Model
    4. Verification
    5. Decision

    Turning image data into measurable information and production-grade decisions through image processing, machine learning and deep learning.

    Conventional image processing where it is sufficient; segmentation, object detection and convolutional networks where it is not; and the deployment and monitoring that make either one production-grade rather than a notebook result.

  • 05

    AI enablement

    A team learning to work with AI, and that capability spreading until several teams build systems of their own.
    1. Understanding
    2. Prompting
    3. Context
    4. Agents
    5. Agentic building

    Helping people progress from understanding AI to using it, designing with it and eventually building systems around it.

    Workshops, structured training and the longer work of building capability inside an organisation — taken in stages, because a team that can prompt well is not yet a team that can build and operate its own systems.