Adam Tyrrell
Work

Systems I designed and built.

Five that are worth reading about: what the problem was, and the decision that made the thing work.

Platform architecture · Team lead150+sites on one platform

Multi-Tenant Site Platform

The shared platform my team ships client sites on: one reusable Next.js theme, a tenant-aware Payload CMS setup, and fully automated pipelines, running 150+ sites on consolidated AWS infrastructure.

I architected it and led the rollout. Consolidating onto shared CloudFront, S3, EC2, and ECR cut infrastructure costs by 90 percent, CloudFormation keeps dev, staging, and production identical, and every deploy passes automated quality and accessibility gates. A new site went from 3-6 months to a single day.

  • Next.js
  • Payload CMS
  • AWS (CloudFront, S3, EC2, ECR)
  • CloudFormation
  • Bitbucket Pipelines
  • Cypress
  • Lighthouse

Built at my day job, so the code stays private. Happy to walk through the architecture.

Production data platform60k rowsbackfilled in ~67s, idempotent

Multi-Tenant Analytics Platform

A multi-tenant analytics platform running in production. Operational reports arrive daily as email attachments, get parsed and imported automatically, and land in isolated per-tenant dashboards.

Every custom API route is tenant-scoped at the boundary, so one tenant can never read another tenant's data. Imports upsert on a stable key, which makes a re-run a no-op instead of a duplicate, and a year of history backfills in about a minute.

  • Next.js 16
  • Payload CMS
  • MongoDB
  • Auth0
  • Express ingest service
  • Recharts
  • Vercel

Private codebase. Architecture and code walkthrough on request.

Automation · AI opsDailyunattended runs

Autonomous Daily Agent Fleet

A fleet of scheduled AI agents that run unattended every day, handling security review, dependency research, and production error watch, then report to a dashboard and push to a phone.

The interesting part is what happens with nobody watching: agents run on systemd timers, a poller survives restarts, and error watch is wired to live Sentry data so a real production exception becomes a notification instead of a surprise.

  • Claude API
  • systemd services
  • Sentry
  • Next.js dashboard
  • Push / email alerts
Multi-agent AI · Data verification522records, each with evidence

Multi-Agent Verification Pipeline

A staged pipeline that runs parallel AI agents to research, verify, and enrich web data, with every claim traced back to firsthand evidence rather than a scrape.

Verification is the whole trick. An early pass read static HTML and reported features as missing that were actually there, rendered by JavaScript, so roughly half of its "this site lacks X" claims were false. Re-checking against the rendered DOM fixed it.

  • Claude API
  • Multi-agent orchestration
  • Node.js
  • Rendered-DOM verification
  • MongoDB
Front-end rangeconcepts

A little front-end range.

Fictional brands, each built to push a small site hard in one direction. Live and clickable.

Happy to go deeper on any of these: the architecture, the tradeoffs, or the parts that did not work the first time.

Get in touch