AI Systems Architect & Product Builder · Founder @ LookADev
I design and build intelligent systems that turn complex business processes into scalable operations.
LookADev · LinkedIn · Case Studies · Email
I take repetitive admin off UK practices and small teams: enquiry capture and routing, booking handoffs, document-to-ERP pipelines, CRMs and dashboards. Deterministic logic wherever a wrong answer is costly, AI only where it earns its place.
In production: a pipeline for Cainelli Bebidas that writes supplier invoices into their Tiny ERP against a 5,222-product catalogue. Early on it created about 60 duplicate products; I traced the root cause, fixed it with a local catalogue index and anti-duplicate locks, and published the post-mortem.
Diagnosis → 7-day sprint → care plan. Start with the free 40-second diagnosis.
⭐ Featured open source: lcc — Local Context Compiler
Shrink what you send to the model. Keep every byte you didn't cut. Get a receipt. −66% tokens on a real Claude Code session with every user/assistant message kept byte for byte. Offline, MIT, CLI + MCP server + Claude Code plugin.
pip install local-context-compiler · 
Most businesses don't have a software problem. They have a systems problem — leads split across four channels, data copy-pasted between WhatsApp, spreadsheets and a CRM, and no one clear on who should act.
I diagnose the real workflow, model the architecture that removes the manual handoffs, and build the system. I sit at the intersection of AI, software engineering, business process and product thinking — which is why I design the solution before writing code, and measure the outcome after.
AI Systems · AI Automation · Business Process Automation · Custom Software · AI Agents · Data Systems · Operational Systems · Product Engineering
Diagnose → Map → Quantify → Architect → Automate → Build → Measure
I never start with technology. I start by understanding the business, mapping the real process, quantifying the cost of the current state, and only then deciding the architecture.
Each of these is a system I designed and built — a business problem turned into an architecture, not a feature list. Full case studies on LookADev.
| System | The problem | The system |
|---|---|---|
| AgentTrace Studio | AI agents are unreliable in real codebases | An evaluation & observability platform that gates AI agents through CI/CD quality checks before they ship |
| LookaBerry | Cold outreach burns time and gets leads ignored | An autonomous GTM system that ranks leads in-database (pgvector), enriches with 0 LLM tokens, and sequences outreach with anti-ban guardrails |
| AI Reception | Small businesses lose leads after the first enquiry | An AI receptionist that captures, classifies and qualifies enquiries, then hands hot leads to a human |
| LookaCrawler | Web pages waste thousands of tokens for AI agents | A token-optimized crawler that strips >73% of HTML noise and exposes a native MCP server |
| lcc | Prompt context is bloated and expensive | A local-first context compiler that cuts prompts and agent sessions (−66% on a real Claude Code session) while keeping every kept byte verbatim, with an audit receipt |
| OpsCommand | Operations data is scattered and unreadable | An accessible operations dashboard that turns scattered KPIs into a single command view |
- LookABerry — closing the feedback loop between outbound signals and intent scoring, so the system gets smarter as it runs.
- AgentTrace Studio — expanding the eval gate so it catches citation and regression failures automatically.
lcc · lookacrawler · agentic-prompt-intake · lookapentest · antigravity-gemini-extension
LookADev · LinkedIn · GitHub · Email