I build systems that connect real business data to AI models and deliver the output into working operations. Currently building and delivering AI operational systems for SMB clients through my own agency.
Most organisations trying to adopt AI have the same gap: they have data, they have access to models, and no reliable system connecting the two in a way that can be trusted and maintained.
I build that system — data ingestion and validation, AI model integration, structured output pipelines, approval gates, and client-facing interfaces — as repeatable, auditable infrastructure.
Current focus: AI integration, pipeline architecture, prompt engineering, model testing and validation, operational knowledge capture, and AI governance for businesses without dedicated data teams.
AI operational system for a restaurant in Mumbles, Wales. Full pipeline built on real POS and bookings data.
- 46,304 POS transaction rows ingested, cleaned, and validated
- 5,000 booking records processed
- Revenue forecasting, contribution margin engine, and no-show scoring live
- 44/44 tests passing on no-show model
- Weekly review pack with operator approval log in active use
- HTML dashboard auto-refreshing every 5 minutes, bookmarked by client
| Layer | Tools |
|---|---|
| AI models | Claude (Anthropic), OpenAI Codex / GPT |
| Language | Python |
| Data ingestion | dlt |
| Schema validation | Pandera |
| Local analytics | DuckDB |
| Workflow automation | n8n (self-hosted, Docker) |
| Task and context hub | Notion |
| Version control | GitHub |
| Client interface | HTML dashboards |
Plan → Delegate → Build → Validate → Review → Approve → Ship
Every task is planned before a line of code is written. Tasks live in Notion with structured prompts — delegated to Claude Code or Codex based on task type and cost. Context flows between models and sessions through CLAUDE.md persistent context files. Every data source has a contract before any logic runs on it. Every model output is tested and confidence-marked. Branch-per-task on GitHub throughout.
Restaurant · Retail · Salon & Beauty · Gym & Fitness · Trades · E-commerce
Full methodology, architecture, and validated client results. Six docs covering:
- 8-stage pipeline design and automation tiers
- Multi-model orchestration (Claude Code, Codex, Notion task hub)
- Operational knowledge capture (kitchen intake, supplier intake, legacy sources)
- Client-facing system architecture (dashboard layers, approval chain, Notion backend)
- Agency scaffold and learning loop (relay system, repeatable templates)
- Per-vertical implementation notes with real Nok Nok delivery data
Open to AI integration, pipeline engineering, and AI operations roles. Available for contract and full-time engagements.

