/work

Problem

Teams struggle to ship AI-enabled systems quickly without losing control. Projects stall in discovery, prototypes don't harden into production, and AI work becomes fragile, opaque, or over-engineered.

Approach

Built Nought.digital as an outcome-driven delivery practice with a fixed execution model.

• Defined a clear end-to-end process (Outline → Research → Build → Integrate → Test)
• Focus on shipping real systems, not decks or demos
• AI used as an accelerator, not a black box
• Emphasis on observability, guardrails, and human-in-the-loop workflows
• Delivery scoped by outcomes, not hours
• Designed to drop into teams, unblock delivery, and leave behind production-ready systems.

Outcome

• Complex projects delivered in weeks instead of months
• Clear ownership, predictable delivery, and fewer rewrites
• AI systems that teams can understand, control, and extend
• Repeatable framework now used across multiple real projects

Problem

Restaurants need to manage allergen data accurately and keep menus up to date. Most systems are static, error-prone, and awkward to update, which creates real legal and health risk.

Approach

Built a SaaS platform focused on speed, structure, and compliance.

• Structured ingredient and allergen model
• Real-time menu updates without redeploys
• Embeddable menus and QR-friendly delivery
• API-first architecture for custom integrations
• Structured workflows to reduce manual mistakes
• Multi-tenant product foundation designed to scale

Outcome

• Restaurants can update menus instantly
• Compliance risk is reduced through structured data and validation
• Menus stay consistent across web, QR, and embeds
• Platform is ready for SaaS scale

Problem

Automotive parts are often compatible across multiple manufacturers and models, but compatibility data is fragmented and difficult to search. People repairing vehicles frequently struggle to determine which parts are interchangeable, which manufacturers produce equivalent components, and whether a cheaper alternative exists. Most databases are incomplete or locked behind proprietary systems.

Approach

Built a platform designed to map equivalent parts across manufacturers.

Key ideas:

• Structured database of part relationships
• Cross-reference search engine
• Compatibility mapping between brands and models
• Scalable architecture to handle large datasets

The goal was to turn scattered compatibility information into a clear and searchable system.

Outcome

• Easier discovery of equivalent parts
• Reduced time spent researching compatibility
• Foundation for a larger automotive parts intelligence platform
• Structured dataset that can grow as new mappings are added

Problem

Resellers often list the same item across multiple platforms such as eBay and Vinted. Managing listings manually across marketplaces is repetitive and error-prone. Common issues include duplicate data entry, inconsistent pricing or descriptions, and listings remaining active after an item sells elsewhere. Existing tools are usually rigid or expensive.

Approach

Designed a platform to simplify and automate cross-listing workflows.

Key ideas:

• Single source of truth for listings
• Automated cross-posting to marketplaces
• Structured item data that can adapt to platform requirements
• Lightweight automation to reduce repetitive listing work

The focus was on reducing friction for resellers, not building another complex listing tool.

Outcome

• Significantly reduced time spent creating listings
• Consistent product data across platforms
• Fewer mistakes caused by manual duplication
• Scalable foundation for marketplace automation

Problem

Many WordPress sites accumulate unnecessary scripts, styles, and plugins over time. This leads to slower page loads, difficult debugging, and fragile performance tuning where fixes often break something else. Most optimisation tools are either overly complex or rely on aggressive caching rather than addressing the root cause.

Approach

Built a lightweight plugin focused on visibility and control over what actually loads on a page.

Key ideas:

• Inspect scripts and styles loaded by WordPress
• Allow selective disabling of assets per page or globally
• Provide safe guardrails so changes do not break the site
• Keep the interface simple enough for developers and power users

The goal was to make performance debugging understandable, not magical.

Outcome

• Faster load times by eliminating unnecessary assets
• Clear visibility into what WordPress is actually loading
• Safer optimisation without brittle configuration
• Developer-friendly tool that solves real performance issues rather than masking them

Problem

WordPress onboarding tools like MerlinWP are theme-bound, brittle, and poorly suited to modern productised workflows. Setup logic lives in themes, making it hard to reuse, version, test, or maintain - especially for agencies and SaaS-style WordPress products.

Approach

Built a plugin-first onboarding framework that decouples setup flows from themes entirely.

• Plugin-based architecture instead of theme drop-ins
• Guided, multi-step setup wizard reusable across projects
• Works with any theme or page builder
• Designed for agencies, product teams, and support workflows
• Clean extension points for custom steps, logic, and validation
• Licensing model aligned with real-world usage (personal, agency, unlimited)
• Focused on separation of concerns, reusability, and long-term maintainability.

Outcome

• Onboarding logic becomes reusable, testable, and upgradable
• Faster project setup without locking behaviour into themes
• Cleaner handover from build to support teams
• Enables SaaS-style WordPress products with predictable setup flows
• Production-ready plugin with pricing and real-world use cases

Problem

WordPress testing is slow, brittle, and hostile to modern workflows. PHPUnit boilerplate, unclear setup, and poor DX mean many teams either avoid tests entirely or maintain fragile, hard-to-read test suites.

Approach

Built a Pest-first testing framework for WordPress that brings modern, expressive testing patterns into a legacy ecosystem without breaking compatibility.

• Native Pest integration tailored for WordPress
• Clean abstractions over WP bootstrap, factories, and globals
• Sensible defaults to reduce setup friction
• CI-friendly configuration for GitHub Actions and GitLab
• Extensive documentation and real-world testing examples (WooCommerce, ACF, roles, caching, background jobs)
• Focused on clarity, speed, and developer confidence, not just test coverage.

Outcome

• WordPress tests become readable, concise, and maintainable
• Faster test authoring with significantly less boilerplate
• Enables teams to adopt modern testing without rewriting their stack
• Improved confidence when refactoring plugins and core logic
• Open-source framework actively usable in real projects

Problem

WordPress search is keyword-based, brittle, and poorly suited to large or content-heavy sites. Important information is hard to find, relevance is low, and extending search quality requires complex custom work.

Approach

Built a semantic retrieval layer for WordPress, designed to improve search relevance without replacing existing content workflows.

• Content indexing and embedding pipeline
• Semantic query matching instead of keyword-only search
• Structured retrieval with relevance scoring
• Works alongside existing WordPress data models
• Designed as a plugin with clear configuration and extensibility
• Focused on retrieval quality, performance, and developer control, not opaque AI behaviour.

Outcome

• Significantly improved content discoverability on large sites
• More relevant search results without manual tagging
• Search behaviour that scales with content growth
• Drop-in infrastructure that does not require rebuilding sites

Problem

Refactoring ACF field structures is risky. Renaming fields, changing layouts, or reorganising flexible content can silently break data, corrupt content, or require fragile one-off scripts. Teams often avoid necessary changes because rollback is painful or impossible.

Approach

Built FieldCraft, a tool for safe, repeatable ACF data migrations.

• Declarative mapping from old to new field structures
• Supports nested, repeater, and flexible content fields
• Dry-run and validation workflows before applying changes
• Reversible migrations to protect production data
• Designed to run safely on live sites
• Focused on data integrity and developer confidence, not speed at all costs.

Outcome

• Enables large ACF refactors without data loss
• Eliminates risky manual scripts and database edits
• Makes long-term schema evolution feasible
• Safer production changes with predictable outcomes

Problem

Craft CMS developers had no polished, Craft-aware local development tool. Setups were manual, inconsistent across teams, Docker configs were hand-rolled, onboarding was slow, and "works on my machine" was the norm.

Approach

Built a dedicated desktop application for Craft CMS development, designed around reproducibility and developer experience.

Key decisions:

• Craft-aware project detection and versioning (Craft 3/4/5)
• Opinionated defaults over endless configuration
• Blueprint system to make environments reproducible and shareable
• Container backend abstraction (Colima, Docker)
• Desktop UI backed by a Go daemon with a clean HTTP API boundary
• Focused on DX first, not just container orchestration.

Outcome

• Local Craft environments spin up in minutes instead of hours
• Teams get consistent, repeatable environments via blueprints
• Local to Craft Cloud workflow unified in a single tool
• Mac-only native app to start, with stable performance and predictable setups