A body of work
I make tools for the inner life, for privacy, for honesty in machines, and for open knowledge about the real world. The thread through all of it is simple: technology should return agency to the person using it — and be built with rigour, not hype.
How I build
I work with AI across the whole lifecycle — one person covering ground that used to need a team.
I care how the pieces fit and hold up over time, not just whether a demo runs once.
Open data, local models, and honest sourcing by default; closed only when there's a real reason.
I let ideas roam, then test them adversarially. Curiosity first, then proof.
The person comes before the platform. If a tool doesn't give someone more agency, it isn't finished.
Tools for the inner life. Most software fights for your attention; these try to give some of it back — helping people think, feel, and reflect more clearly.
A binaural / Hemi-Sync-style Android app for guided cognitive states — meditation, focus, and altered attention through carefully generated audio. Built natively for reliable, low-latency sound on real devices, not a web wrapper.
A multi-perspective inquiry engine that examines a question through many systems of thought at once — scientific, ethical, strategic, creative, contemplative — so an answer is stress-tested from every angle instead of one.
An exploratory line of research mapping emotional states onto harmonic ratios — asking whether the structure of how we feel has a describable, musical geometry. Early-stage, held as an open question.
Technology that returns control to people. As digital-ID, KYC and age-assurance quietly become mandatory, the right to move through the world without being catalogued is worth defending — lawfully, and by design.
Privacy and digital-sovereignty tooling that detects where a service is imposing digital-ID, KYC or age-assurance checks, and maps the lawful alternative routes a person can still take. It tracks the regulatory landscape (Online Safety Act, One Login, eIDAS 2.0) and helps people understand and reclaim their own data footprint.
AI that tells the truth and belongs to everyone. Models are trained to please, and increasingly owned by a few. These projects push the other way — toward honesty, self-scrutiny, and open access.
An anti-sycophancy harness. It deliberately collides distinct perspectives against an idea, then adversarially verifies what survives — with web-checked claims rather than flattery. Built because the most dangerous thing an AI can do is agree with you convincingly.
Automated discovery, licence-vetting and curation of open local LLMs — pulling weights, model cards, licences, provenance and checksums into a self-hosted library. Democratising access so capable AI doesn't only live behind someone else's API.
A routing layer that spends intelligently across frontier and local models under hard cost caps — with graceful fallback when one model refuses or falters. Keeps AI use affordable, private-by-preference, and never locked to a single vendor.
An independent commentator that watches the system it lives in and critiques its own behaviour — flagging when it's talking too much versus staying too silent. A small experiment in giving an AI a genuinely separate, self-questioning voice rather than a compliant one.
Open reference data for real-world systems. Good decisions about land and food need good, public information. These build the maps and knowledge layers that everyone can use.
A public directory that maps the UK's farm shops and local producers — a 223,000-farm database, 99.5% geocoded, served fast from Cloudflare's edge with an interactive vector map. Open reference data that helps people find food closer to where it's grown.
A UK land-analysis desktop tool — a native app for exploring and reasoning about land: parcels, constraints and characteristics, brought together to support clearer, better-informed decisions.
A Cognee-powered knowledge layer that turns scattered documents and notes into a connected, searchable graph — so information can be reasoned over as a whole rather than sitting in unread folders.
Being honest about what AI actually displaces. Most of the debate is either denial or hype. This work measures it instead — sector by sector, against the regulators' own published record — and then builds the tooling the affected professions will need.
A framework that scores 40 UK professional sectors on how exposed they are to being disintermediated by AI — eight weighted dimensions spanning vulnerability, resistance and market opportunity, from financial auditing at 75/100 down to data entry at 35. The scores aren't left to go stale: it ingests live from official primary sources — the FCA, HMRC, HM Land Registry and the Property Ombudsman — and flags the changes that should move a sector's score.
The second half is what follows from the analysis. For each scored sector it generates the compliance documentation those firms are obliged to hold — 156 practice areas, 853 checklist items, and packs running to hundreds of pages (AML policies, UK GDPR, complaints handling, SRA and FCA manuals) — with a client portal, filing-deadline tracking and a full audit trail. If a profession's routine work is going to be automated, the honest response is to build the thing that does it properly, not to pretend it won't happen.
If any of this resonates — as a collaborator, a fellow builder, or someone who just wants to talk about it — I'd genuinely like to hear from you.