I design and ship multi-agent systems, voice AI, and RAG pipelines for regulated and operationally sensitive industries: healthcare admin, professional services, B2B SaaS. The wedge is three years auditing a $1B company's Oracle Digital Transformation rollout at PwC A.F. Ferguson before I built my first agent. Most AI engineers see a tech problem. I see the business process underneath.
Five flagship builds, each shipped, each with verifiable architecture and outcome metrics. Plus two interactive demos you can try right now.
Agencies running LinkedIn outreach across many client accounts hit the same wall: calendar-based rate limits cause silent overspend at midnight, single LinkedIn accounts get restricted, and there is no consolidated dashboard for invites accepted, messages sent, and acceptance-rate trends across the book of business.
Existing LinkedIn automation tools either get accounts restricted (no safety guardrails) or feel like spammy form-fillers (no real personalisation). What was needed: AI-personalised outreach that respects LinkedIn's safe usage limits absolutely and gets nothing about the prospect wrong.
Clinical practices and service businesses cannot scale content production manually. Hiring an in-house team is expensive. Generative AI alone hallucinates compliance-sensitive details. What was needed: a HITL pipeline that lets a non-technical operator generate brand-aligned video and image content from a single Telegram command, with cost guardrails.
UK aesthetic and dental clinics lose 15+ minutes per patient on manual data entry from paper, PDF, photo, or typed intake forms into Pabau or Cliniko EMR systems. Existing OCR tools require human review on every field. Generative AI alone cannot be trusted on contraindications. What was needed: a GDPR-compliant pipeline that automates the safe cases and routes the risky ones to a human.
Two live, interactive demos. No login. Click and use.
LIVE
Interactive portfolio with an embedded live VAPI voice agent. Three.js snowfall, WebRTC streaming, sub-1200ms response. Recruiters can dial in and have a conversation.
LIVE
Monorepo showcase: React Router v7 web app plus Expo SDK 53 mobile app, both wired to a live n8n chatbot webhook. Cross-platform, dark mode, animated, mobile-ready.
Pick the entry point that matches where you are. Pricing is shaped by scope and discussed once we've talked. Never out-of-the-box rate cards.
Custom AI builds, architecture audits, fractional engagements, exploratory conversations. A 15-minute call is the fastest way to scope it. No prep needed. Bring the problem.
Most AI engineers see a technology problem. I see the business process underneath: the manual steps, the exception handling, the audit trail, the part that breaks when the system scales from 10 users to 10,000.
That precision was forged over three years at PwC A.F. Ferguson & Co. (ACCA-qualified). I served as a techno-functional consultant on a $1B company's Oracle Digital Transformation rollout, led a cross-functional team of eight, mapped multi-stage manufacturing workflows in strict IFRS adherence, and engineered systems supporting 10,000+ daily inventory movements at sub-second latency.
I now design and ship production AI. The flagship is a markdown-driven AI Operating System orchestrating eight parallel sub-agents (~50,000 lines of TypeScript), the system that self root-caused Anthropic Claude Code Issue #9458 to operate reliably. Beyond the OS: a multi-tenant SaaS outreach platform, a security-first MV3 Chrome extension with 8-layer safety architecture, a Telegram-driven generative content factory with 97% test coverage, and an EU GDPR-compliant clinical intake pipeline that compresses 15-minute manual workflows into 3 seconds.
The throughline is process discipline before the model. Schema before prompt. Audit log before agent. Cost ceiling before vendor call. Production AI lives or dies on these decisions long before the LLM gets called.
I am currently pursuing Anthropic's Claude Certified Architect (CCA-F), the first proctored credential for the Claude / MCP ecosystem, and AWS Solutions Architect Associate. Stack spans Claude API, the Anthropic Agent SDK, MCP, TypeScript, Python (FastAPI), Supabase, Cloudflare Workers, and the VAPI / Retell / Eleven Labs voice ecosystem.
If you need a system that survives production, not just a demo, let's talk.
Six current certifications across cloud, data, AI, and process automation. Three more in progress targeting the AI architect tier.
The fastest first step is a 15-minute call. We'll talk about what you're building, where you're stuck, and whether a Technical Assessment is the right next move for both of us.