I build the systems
the work comes out of.

A multi-agent system that runs my own job search

It queries the Greenhouse, Ashby, Lever and Workday APIs directly across ninety-plus companies, scores every requisition against a weighted rubric, and kills anything that fails a hard gate before it reaches me: wrong country, a listing that is already dead, a compensation floor, or a job title that means something different from what it appears to mean. Agents have defined roles. State moves between files and nothing is ever deleted, so every decision has an audit trail I can go back through.

I approve everything. It decides what is worth my attention. That division is the whole design, and it is the same one I would build for a team.

The production pipeline and prompt optimizer

A 46-node pipeline spanning seven-plus AI platforms, taking script to storyboard to high-resolution first and last frames to finished clips with audio. Alongside it, a prompt optimizer that reads the generated storyboard back against the original script treatment and writes the timestamped direction between keyframes, so camera and performance intent survives into the shot instead of dying at the prompt. Clip generation went from hours per scene to fifteen minutes.

Knowledge architecture for AI systems

Built from the ground up for small business owners and corporate marketing teams, starting with custom GPTs. The work is not prompting. It is structuring, curating and pruning the source material until retrieval returns the right thing rather than the nearest thing.

These systems fail on thin or contradictory source documentation, not on model choice, so the real deliverable is the exhaustive requirement set underneath. That is the difference between something that demos well and something that survives daily use by an operator who is not technical.

The governance nobody wants to write

In most organizations the blocker on shipping generated work is not craft and it is not model quality. It is that nobody has an agreed answer on what discloses, who reviews, and which step it replaces. So the work is good and stuck. I wrote the framework that governs customer-facing generated content, plus 86 pages of production standards across eight sections with review gates and human-in-the-loop checkpoints written in, and trained the team against it.

Hubs a production can actually run out of

A per-project video production hub carrying goals, tasks, file management, prompt libraries, first and last frame holders, storyboarding and shot-by-shot tracking in one place. A blog pipeline hub for status, review and editing across a full calendar. Multiple websites and design system hubs. A task manager and command center built to match how I actually work rather than how project software assumes I do.

Two years of AI-assisted building across Claude Code, Cursor, Replit and Lovable, including connecting MCP servers and APIs. I am not a software engineer and do not claim to be. The skill is specifying precisely enough that the thing gets built, and then knowing whether it works.

This page

Hand-built, version controlled, custom domain, live about thirty seconds after a push. It is the smallest thing on this list and it makes the same point as the rest: I would rather ship the thing than commission it and describe it.

AI Film Studio: proposed, proven, adopted

Proposed the AI video production operating model to leadership, was assigned to lead it, and proved it across two proofs of concept. The first was a complete 35-shot explainer directed end to end against a ten working day deadline, with a team spread across France and Denver: a graphic designer, a videographer and editor, and a video production graduate. None of them had used any of the tools, two were French native speakers, and all three arrived with very different appetites for change. The second I built independently and took to leadership as a business case. Both were approved, and the processes and tools now run all Harmonic video storytelling, external and internal.

Launch films for the biggest product launch

Seamless single-shot animated pieces running thirty seconds to two minutes. Transitions carried inside the shot, no cuts, narration integrated, the whole thing held inside a controlled brand style. Produced on a tool I introduced after building the proof independently and taking it to leadership as a business case: lower production cost, faster turnaround, less headcount required per production, and storytelling range the previous process could not reach.

Visual identity and the agent character system

Two years of disparate sub-brand assets consolidated into one visual identity system, built after the design function was vacated and without allocated budget. Includes a family of seven AI avatar characters, one for each vertical of the new AI product, sharing a design language, colour coding and holographic staging carried across launch, product and campaign surfaces. Goes live this month.

Neuroflix: original vertical series

Vertical micro-series development for neurodivergent storytelling across comedy and drama. Characters, sets, scripts, visual language and series bibles, written, designed, directed and produced solo. Run as a production system rather than a set of one-off pieces, which is what makes output at this scope and pace possible.

Client and consulting work

Sixteen years of brand and creative strategy: brand narrative, visual identity, high-conversion landing pages, long-form content and video scripts, with creative direction for multimedia production. Hired, directed and developed freelance designers, illustrators and motion people, and managed agency and studio partners.