For hiring teams
Forward-deployed AI builder
I embed with domain teams, find the manual work, and ship the agent that removes it. 15 years in physical product development, 4+ years shipping AI.
Proof
Proof tracks, not a project archive.
The signal is not that I built many things. It is that the same operating model shows up across professional workflows, physical systems, and revenue products.
AI workflow systems
I turn recurring expert work into command centers, scoring loops, and AI-assisted operating rhythms.
The Wire
Automated daily product-intelligence feed — sourced, scored, and published by an agentic pipeline.
Stakeholder Lens
Paste a roadmap once and generate audience-specific views for engineering, sales, execs, and customers.
Implementation Scoping Tool
Describe a business problem, answer discovery questions, and receive a structured AI implementation scope.
Industrial and physical systems
I apply agentic workflow thinking to manufacturing, CAD, product-development, and operations problems.
Forkable Factory Map
Public map of the 10 product-creation phases and how far each can already be made forkable with agents.
Photo-to-Print Benchmark
Drop a photo, generate printable STL candidates, and score image-to-3D models on printability.
Agent CAD Probe
Weekly benchmark of AI-native CAD tools, scored on agent-readiness, evidence quality, and workflow maturity.
Founder and revenue products
I have shipped products beyond demos: AI pipelines, fulfilment, customer-facing experiences, and revenue.
Personalised Children’s Book Platform
AI-generated personalised stories and illustrated pages routed through print-on-demand fulfilment.
Downtime Cost Calculator
Manufacturing ops calculator that turns incident data into downtime cost, annual risk, and monitoring ROI.
Roman's Lab
The wider collection of PM tools, AI workflow prototypes, manufacturing experiments, and shipped utilities.
Method
How I build AI operating systems.
The craft is not prompting. It is turning expert work into a system with memory, cadence, judgment, and review.
Map the real workflow
Start with the human operating rhythm: inputs, handoffs, judgment calls, bottlenecks, and failure modes.
Find the leverage points
Separate where AI should decide, draft, retrieve, critique, summarize, or simply stay out of the way.
Design the agent loops
Turn recurring work into repeatable commands, prompts, trackers, and review loops that can be run every week.
Build the workspace
Claude Code, Markdown, Git, APIs, and lightweight UI where useful. The system should be usable, inspectable, and adaptable.
Measure execution
Track the work that matters: applications, outreach, shipped artifacts, decisions, blockers, and next actions.
About
Operator roots. AI hands.
I started in physical systems: factories, industrial equipment, global engineering teams. Then I spent years building AI products and agentic workflows. The through-line is turning messy expert work into operating systems people can actually use.
Leading a 10-person cross-functional product development department. Responsible for new product development and lifecycle upgrades of complex industrial equipment across Europe and China.
Led product strategy for a personalised AI-powered publishing platform. Built end-to-end AI workflows, prompt architectures, and QA systems – scaling to 1,000+ books produced. Also ran an AI automation agency building custom GPT-based systems for business clients.
Led cross-functional engineering teams delivering complex industrial and manufacturing solutions for global clients, including additive-manufacturing initiatives.
Grew from injection molding technician to Lead Professional in LEGO's Concept Factory. Led manufacturing innovation initiatives and coordinated global engineering teams.
How I work
Tools & stack
Contact