Work
Every project below is broken down the way I'd brief a board: the situation, the problem, the architecture, the stack, what shipped, and what it proves. Read them, steal the patterns, run them yourself. That's the point. Each one links to the playbook that teaches the pattern behind it.
Each entry is labeled honestly: client delivery, an operated P&L, internal R&D, a pilot, or a shipped product. No inflated claims, no projections presented as results.
At Yelp I co-designed the infrastructure that moved a 30+ PB/month data pipeline onto Kubernetes with a team of five, cutting several thousand dollars a day in cost while keeping storage, ML, backups, and customer reporting running.
30+ PB/month processed; several thousand dollars/day saved; a workflow state manager that returned hundreds of engineering hours a month
At Index Exchange I built an ML-enhanced rate-limiting model that lifted server performance 30% across a 5,000-server global fleet while holding 95% of revenue, and found a $10,000-a-month cost reduction in the infrastructure data.
30% performance gain at 95% revenue retention across 5,000 servers; $10,000/month cost reduction identified and acted on
On contract at a consumer social platform with 100M+ users, I architected the centralized LLM orchestration that put premium AI features into production, migrated models to clear a 40M-event backlog, and chartered the internal AI Guild.
Premium AI features in production for 100M+ users; a 40M-event backlog cleared on a model migration; an internal AI Guild chartered
I operated a real battery ecommerce P&L on eBay, $29,138 net over five months, then built the owned WooCommerce channel to escape marketplace rent.
$43,643 gross / $29,138 net (Jan–May 2026); 313 sales at 100% positive feedback
An IT-services company ran on manual call handling and a spreadsheet catalog. I deployed a bidirectional voice-AI phone agent connected to a self-hosted CRM, on their own infrastructure.
Voice agent + CRM + catalog sync deployed to production on the client's VPS
A generative asset pipeline where every run writes to a cost ledger: 256 production assets delivered for $37.59, about 15 cents each, with the receipts to prove it.
256 production assets for $37.59 (~$0.15 each) with a per-run audit ledger
Genetic testing labs ship results as massive PDFs. I built the extraction pipeline that turns them into structured, queryable product data, live against three lab formats.
238 markers extracted from a 170-page report; 3 live lab-format integrations; 21,297-item reference database
A health podcast was losing hours per episode to manual transcription, blogging, and clip-cutting. I built the pipeline that automates the whole chain, with a 12-tab admin UI the owner controls.
Episode-to-blog-to-clips pipeline live; 6 episodes processed; phases 1–5 of 8 delivered
Three integrations built for owner-led businesses: a restaurant's thermal-printer order bridge, a SharePoint-to-Square catalog sync running in production, and an AI invoice extractor. Unglamorous, high-leverage.
3 back-office bridges built for 3 real businesses; the catalog sync runs in production
Five apps live across the App Store and Google Play, not as a portfolio brag, but as a repeatable shipping methodology: automated release pipelines, store compliance, monetization wiring, reused every time.
5 apps shipped to public stores on one reusable release pipeline
Five WordPress properties for real clients, running on a self-managed VPS behind Traefik with CI/CD, snapshot-safe deploys, and written runbooks, operated with postmortem discipline.
5 live client properties; documented runbooks; incident-to-postmortem operating culture