Problem
No independent evidence on whether Scotland's AI strategy is being delivered.
Problems I solve
If any of this sounds familiar, we should talk.
These are the recurring, expensive, hard-to-scale problems I take on. If one is costing your team time or money, that's the conversation to start.
📄
Repetitive admin
Email, documents, data entry and routing that eats hours every week.
⚠️
Unreliable AI
Demos that work and production that doesn't. Hallucinations, drift, no guardrails.
🔍
Knowledge retrieval
Answers buried in documents your team can't search or trust.
🔗
Integrations
AI that doesn't connect to the CRM, inbox or systems you already run.
🤖
Agents & automation
Workflows that need orchestration, approval gates and audit trails.
🛡️
Governance & control
Who can the AI touch, and what happens when it's wrong?
Services & pricing
AI systems built for reality. Not just the demo.
Design, implementation and hands-on technical ownership for UK organisations that want AI to remove work, improve decisions and connect to the systems they already use. Fixed-scope implementation first — larger programmes only scale into dedicated delivery after the economics and the need are real.
AI Opportunity & Automation Audit
Process review, data readiness, prioritised opportunity map, 90-day roadmap.
£1,500–£2,500 · 1–2 weeksAI Workflow Automation
Email/document intake, classification, decision routing, CRM updates, human approval steps.
£5,000–£12,000 · 2–4 weeksInternal AI Knowledge System / RAG
Document ingestion, retrieval architecture, source-grounded answers, permissions, citations.
£7,500–£15,000 · 3–6 weeksAI Sales & CRM Intelligence
Lead health signals, meeting prep briefs, CRM summarisation, opportunity risk alerts, management digests.
£8,000–£18,000 · 3–6 weeksAgentic Operations System
Tool/API integration, state orchestration, approval gates, least-privilege access, audit trail.
£12,000–£30,000 · 4–8 weeksAI MVP Productionisation / Rescue
Architecture review, security/secrets review, error handling, observability, deployment hardening.
£5,000–£20,000 · 2–6 weeksAI Governance, Permissions & Control Review
Data access mapping, role/permission review, human approval points, agent tool authority, audit/logging.
£4,000–£10,000 · 2–4 weeksBespoke AI Platform / Integration
Custom build: discovery, architecture, web/backend engineering, AI/ML components, cloud deployment, data pipelines, QA, handover.
From £20,000 · scope-ledIndicative prices are subject to discovery, scope, delivery risk and requirements. Third-party licences, model/API usage and cloud costs are excluded unless specifically included. AI systems are probabilistic; where consequences matter, appropriate human review and control should remain part of the design.
Selected work
Built, tested and published.
Every build publishes the problem, the stack, the result and the failure modes. No manufactured certainty.
Client Systems
Client System
Scotland AI Observatory
Built
Evidence-led observatory tracking all 10 national AI commitments with sources and dates.
Result
10 commitments tracked · adoption index · SME diagnostic
Stack
Cloudflare Pages · Workers · D1
Open Source
Open Source
AgentInterdict
Problem
Autonomous AI agents can take unsafe actions with no runtime enforcement.
Built
Runtime enforcement engine: trust context, verify authority, interdict unsafe action.
Result
96.5% block rate on a 200-attempt injection suite
Stack
Python · Cloudflare Workers · D1
Open Source
SovereignRoot
Problem
AI agents have no human-owned boundary they cannot cross.
Built
Signed sovereignty.json constraint policy, local-first and vendor-neutral.
Result
Free · private · cryptographically signed
Stack
Static · local-first
Open Source
Assentia
Problem
Browser agents silently grant more consent than their user delegated on cookie and tracking interfaces.
Built
Local delegated-consent enforcement layer giving agents a policy decision point before consent UI.
Result
Fail-closed · zero-network by default · MIT
Stack
Python · FastMCP · optional local Ollama
Open Source
CatalogLock
Problem
Agents discover MCP and ARD catalogs, then connect blindly. Stable identity is not stable capability; reviewed state drifts.
Built
Deterministic pre-connect security gate: validates ARD entry sources, snapshots MCP surfaces under explicit profiles, freezes reviewed state into a CI-diffable lockfile.
Result
135/135 tests passing · zero runtime dependencies · flags critical drift like a newly added delete_invoice tool
Stack
TypeScript · GitHub Action · SARIF output
Build Lab
Build Lab
ShieldedID
Problem
Identity systems store PII that becomes a breach target.
Built
Zero-knowledge identity protocol with pairwise subject IDs.
Result
Zero PII storage · Bulletproofs ZK in WASM
Stack
WASM · ZK proofs
Build Lab
STCL
Problem
LLM context windows fill up with redundant tokens, raising cost and latency.
Built
Semantic Token Compression Layer, provider-agnostic pre-inference compression.
Result
~38.2% average token cut · ~3.5ms overhead
Stack
Provider-agnostic
Build Lab
UnitedLines
Problem
Social networks are not privacy-preserving by design.
Built
Encrypted social network as a working build.
Result
Privacy-preserving communication
Stack
Encryption-first
Capabilities
What I can build
Hands-on applied AI and full-stack engineering. I build working systems end-to-end — from local LLM orchestration to production web apps on the Cloudflare free tier — and I publish the method, cost and failure modes.
Applied AI & LLM systems
Local LLM orchestration (Ollama), model selection, prompt engineering, RAG and vector retrieval, agent workflows, semantic compression. AI that runs on your own hardware — no vendor lock-in.
Full-stack web & WebGL
Static-first sites on Cloudflare Pages, serverless Functions, Three.js/WebGL immersive scenes, progressive enhancement, SEO, accessibility and performance budgets. This site and the Observatory are built this way.
Automation & data pipelines
n8n workflows, scheduled cron jobs, data ingestion and ETL, vector databases (Chroma), Redis, email automation via SMTP, analytics and conversion tracking. Systems that run themselves.
Privacy-preserving engineering
Zero-knowledge proofs (Bulletproofs in WASM), pairwise identifiers, zero-PII storage, encrypted communication. Privacy is a design requirement, not an add-on.
Technical stack
Python
JavaScript / TypeScript
Three.js / WebGL
HTML / CSS
Cloudflare Pages + Workers
Serverless Functions
Ollama / local LLMs
RAG / vector search
Chroma / vector DB
Redis
Docker
n8n automation
Git / CI
SMTP / email delivery
Turnstile / bot protection
Zero-knowledge proofs
WASM
OSINT tooling
Data pipelines
Scheduled automation
How I work
Evidence over claims
Every build publishes the problem, baseline, stack, time, cost, result and failure modes.
No manufactured certainty
I test claims, including my own, and report limitations honestly.
Cost-disciplined engineering
I prove the workload cheaply on free tiers and local hardware first, and only scale infrastructure when the economics justify it.
Authority & proof
Track what Scotland promises. Measure what it delivers.
I run an independent, evidence-led observatory for Scottish AI: national strategy delivery, adoption data and public-sector progress. All 10 commitments in Scotland's AI Strategy are tracked against observable evidence with sources and dates.
AI Scotland — FAQ
Implementing AI into your business, answered.
Straight answers on cost, scope, safety and ownership. No jargon, no manufactured certainty.
How much does it cost to implement AI into a business?
AI implementation in Scotland typically starts at £1,500 for an opportunity and automation audit, £5,000–£12,000 for workflow automation, and £7,500–£15,000 for an internal RAG knowledge system. Larger agentic systems and bespoke platforms run £12,000–£30,000+. Every engagement is fixed-scope and evidence-led.
What AI services are available for Scottish businesses?
AI services for Scottish business include opportunity and automation audits, workflow automation, internal knowledge systems (RAG), sales and CRM intelligence, agentic operations systems, AI governance reviews, and bespoke AI platform integration. All are delivered fixed-scope from Perth, Scotland.
How do I implement AI into my business?
Implementing AI into a business starts with an opportunity and automation audit: process review, data readiness, and a prioritised 90-day roadmap. Then a fixed-scope build — workflow automation, a knowledge system, or an agentic system — with human approval gates and audit trails where consequences matter.
Can AI run on my own hardware without vendor lock-in?
Yes. I build AI that runs on your own hardware using local LLM orchestration (Ollama), so you avoid vendor lock-in and recurring API costs. This is part of a cost-disciplined approach: prove the workload cheaply first, then scale infrastructure only when the economics justify it.
Is AI implementation safe and governed?
Yes. Every build includes appropriate human review and control where consequences matter: data access mapping, role and permission review, human approval points, agent tool authority, and audit/logging. AI systems are probabilistic, so governance is part of the design, not an add-on.