GPT
Research, synthesis and specification
Useful for reconstructing professional context, challenging interpretations and shaping evidence into clear operating requirements.
AI working practice
AI became useful when it became part of a governed working method.
My use of AI developed from conversational assistance into a working environment for research, evidence analysis, software development, workflow design and controlled autonomous execution.
Development path
From conversation to controlled execution.
- 01
Conversational use
Exploring questions, drafting and learning how model outputs behave.
- 02
Structured research
Giving research tasks explicit sources, objectives and evaluation criteria.
- 03
Document intelligence
Turning large professional archives into structured, reviewable evidence.
- 04
Multi-model work
Assigning different systems distinct roles instead of treating them as interchangeable chat windows.
- 05
AI-assisted development
Moving from ideas into repositories, interfaces, tests and working prototypes.
- 06
Local agent execution
Using Codex with real files, browsers, databases and controlled automation.
- 07
Bounded autonomous workflows
Defining what an agent may execute independently and where human judgement must stop it.
Multi-model working
Different systems. Different roles.
Claude
Long-context review and alternative reasoning
Used when a second analytical perspective or sustained review of complex material improves the result.
Codex
Implementation, testing and controlled execution
Works with the local filesystem, code, browsers and tests to turn an accepted specification into a working system.
VS Code and Git
Human-controlled development environment
Provide the repository, review points, version history and ability to accept, reject or reverse changes.
From prompting to system design
An instruction becomes useful when it defines how work should be governed.
Select a layer to see the operating question and a sanitized CareerAgent example.
Context
What environment and prior state must the system understand?
CareerAgent example
CareerAgent continues from a validated evidence architecture and preserves validated prior work.
Flagship AI case
CareerAgent
Turning a professional archive into an operational career intelligence system
Traditional CVs compressed years of work into titles and short task lists, leaving much of the actual professional depth invisible.
- 01Professional archive
- 02Document extraction
- 03Evidence
- 04Claim validation
- 05Capability mapping
- 06Role matching
- 07Tailored application
- 08Outcome tracking
AI was used to reconstruct professional experience from evidence, not to invent professional experience.
98,560 source metadata records indexed and validated
59 application-safe claims at the latest validated checkpoint
32 evidence-backed interview cases
79 regression tests passing before this public-profile build
AI quality control
The model does not decide when its work is good enough.
A weak result is not automatically solved by making it look better. Sometimes the evidence, workflow or specification must change before the output deserves to be accepted.
- 01Generate
- 02Challenge
- 03Verify
- 04Test
- 05Accept or reject
Selected digital work
DOPRI
Substantial working prototype
An AI-assisted product-development exercise taken through implementation and technical validation.
Technical validation
- 32 tests passed
- Type checking passed
- Production build passed
- Prerendering passed
Transparency
This profile was developed through an evidence-backed AI working process. AI supported analysis, synthesis, implementation and testing. Human judgement determined what was accurate, appropriate and useful to publish.
The purpose is not to automate judgement away. It is to give professional judgement better evidence, stronger tools and a more reliable execution environment.
Download CV