Serverless architecture
A static front end on Vercel's CDN with serverless functions as the API layer — no persistent backend, spinning up on demand. Minimal cost and operational burden, able to scale.
Solution Architecture · AI Systems
Building with AI
I conceive the system, choose its architecture, build it, and ship it.
A senior solution-architect practice for AI-integrated systems — combining three capabilities usually split across three people: system architecture, full-stack engineering, and applied large-language-model work, held together by the product judgment to decide what is worth building in the first place. The value is in judgment and integration, not in volume of code.
01
End-to-end ownership
A solution architect at this level does not implement a specification handed down by someone else. The same system is conceived, architected, built, deployed and documented as one piece of work — making the trade-offs between cost, scale, maintainability and risk along the way. The seniority that lets the work move fast is what makes the architectural decisions sound.
Serverless & cloud architecture
Full-stack web & design systems
Graph data modelling (Neo4j)
LLM prompt engineering — prose → reliable structured data
Service integration into a coherent whole
Deployment & lightweight DevOps
Privacy, compliance & governance — incl. data about minors
02
Case study — the NetSciEd Challenge
Flagship build · network science education
challenge.networkliteracy.org
A global-scale student competition built end to end, in which the subject matter — networks — is embodied in the infrastructure itself: every submission becomes a node in a graph database, so the system is a working demonstration of the network science it teaches. Conceived, built, tested and deployed in a concentrated three-day sprint, then hardened, documented and governed since.
AI classification pipeline
Students write free-text answers; they never pick from a menu. The Claude API reads the prose and returns structured JSON — a network name, the best-fit domain of nine, and the relevant Network Literacy concepts — with retry and manual-review fallback.
Graph database · Neo4j
Five node types — Submission, Concept, Domain, School, GradeLevel — with constraints, indexes and seeded concept and domain nodes. "Every submission sharing two or more concepts" is a traversal, not a join.
Front end & design system
Six pages on one cohesive design system — a defined type scale, palette and navigation — with an eighteen-language mission statement, no framework and no build step, so it loads fast and caches trivially.
Live visualisations
Neovis.js renders submissions as an interactive force-directed graph colour-coded by domain; Leaflet maps each geocoded school; a dashboard aggregates by domain, concept, country and school.
Moderation, safety & privacy
Inbox-based approval holds any uploaded AI creation pending until reviewed. The form is COPPA-aligned — a responsible-adult contact, consolidated consent and IP — with a deliberate USA-only pilot scope.
Powered by Neo4j AuraDB & the Claude API · orchestrated submission flow with atomic, rolling-back transactions · Nominatim · Resend · GitHub → Vercel continuous deployment
03
The CoExplorer ecosystem
A connected family of AI-aware, graph-native learning systems, built on the College of Exploration's three decades of online practice — the line that runs Caucus (1991) → Confabula (2025) → learning with AI (2026).
↗confabula.net
Confabula
The substrate — a graph-powered (Neo4j) threaded-discussion platform on Python & FastAPI, carrying the 50-year Confer/Caucus conferencing lineage into something AI-aware.
↗coexplorer.comCoExplorer (commercial)
AI-guided voice conversations that capture the expertise your best people never wrote down — and turn it into navigable, teachable knowledge. The consulting offer.
↗coexplorer.netCoExplorer Network
The hub that holds the system together — Confabula, Augur, Wayfinder and Workbench, with the subject spaces and applied frameworks.
↗coexplorer.orgCoExplorer Project
The learning modules and subject spaces — Cybernetics and beyond — written in the teachback tradition of Pask's Conversation Theory.
↗coexplorer.ukCoExplorer UK
Browser-based tools — a Concept Mapper that exports to Neo4j, and a Learning Explorer built on Paskian entailment meshes. A project of CoExploration Limited.
↗networkliteracy.orgNetSciEd Challenge
The flagship build above — the network-science competition whose infrastructure is a network.
Tools across the system: Wayfinder (adaptive, learner-aware learning) · Workbench (expert knowledge → learnable structure) · Augur (reading the pattern of collective reasoning) · Concept Mapper.
Knowledge loss, or a system worth building?
If your organisation is losing expertise as people retire — or you want an AI-integrated learning or knowledge system designed and built end to end — let's talk.