Instead of a five-minute screen recording per application, each system I've shipped gets one short film: the problem, how the architecture works, and the measured result. Together they cover most of what agentic AI looks like in production today, from data pipelines to retrieval, generation, computer vision, autonomous engineering and conversational ops. Every brand and conversation on screen is fictional; every number comes from the real system.
1:08
Watch film
Eight sales and ad platforms pulled nightly into one warehouse, a metrics layer on top, and a projection engine that runs thousands of simulated futures into a band the finance team plans against. It replaced a weekly copy-and-paste spreadsheet without changing the sheet finance already used.
0:52
Watch film
An agent pulls the best-performing ad with its real spend, reads it like a strategist, writes a brief for every angle and generates new creative at scale. A brand gate checks logo, colours, claims and legibility, and anything that fails regenerates with the reason attached.
0:58
Watch film
A ticket goes in and agents ship it: a router picks the repo, a builder works in an isolated worktree, integration and security reviewers hold a veto, deploys roll back on a failed health check, and a QA agent tests the live app. 600+ tasks across 16 repos for about $1k of model spend.
0:57
Watch film
Call transcripts, Looms and Slack history chunked and embedded nightly into Postgres, searched by meaning and by exact words with fused rankings, and served through an MCP server so Claude answers in Slack with the call it came from. Built in about two days.
0:43
Watch film
Customers order the way they always have: voice notes, photos, "the usual". The bot turns each message into order lines matched to that customer's catalogue and prices, asks one question when something is ambiguous, and posts the confirmed order to SAP. 99.8% of orders confirm with no human touch.
0:58
Watch film
Raw creator videos come in from four sources. The pipeline extends squares to vertical, tags every clip in 22 dimensions with vision models, replaces burned-in captions, scores music, cuts shorts and QAs every render before delivery. Editing went from about 4 hours to under 8 minutes per ad: 18,000 creatives in, 35,000 ads out.
Open to senior and lead AI engineering roles with enterprise teams and funded startups solving interesting problems with AI.