Custom Ai apps and workflow automation, case studies from this year

A multi-troop program runs on a stack of unrelated SaaS subscriptions. One for registration, another for messaging, a third for trip planning, plus a Facebook group and three Google Sheets that nobody owns. Around $400 a month per program, four logins, and key data living in three places that drift out of sync.

Multi-tenant React + Express + Postgres. Claude Opus is the workhorse for reasoning-heavy work (trip planning, lesson sequencing); Sonnet handles the everyday flow; Haiku covers high-volume classification. The right model for the job, not loyalty to one. Google OAuth, four-tier role-based access. CSRF, CSP, HSTS. Per-tenant token caps with cost monitoring built in from day one. Backups verified nightly by restore-and-diff against production.

Replaces three separate SaaS subscriptions per tenant. Runs at roughly $8 a month per program in infrastructure, the cost of a streaming subscription. Live at traillog.ai, used by real Scout programs in production today. Tri-model orchestration is the pattern that graduated out of Zero Drift into a shipped product.

A one-person shop runs more services than one person can watch by hand. Every dependency (Claude API, n8n, Postgres, Docker images, dozens of CVE feeds) has its own retirement schedule, advisory list, and release cadence. Without a single screen, the failure mode is hearing about a problem from a customer instead of from a monitor.

FastAPI + React + Postgres. 28 modules sweep every dependency on each refresh: uptime probes, model-deprecation tracker, CVE cross-check against my SBOM, per-tenant token-spend rollup, backup-verify that restores yesterday's backup to a throwaway database and diffs it against production. One Opus 4.7 call synthesizes the morning brief. Public read-only view strips internal pages and customer data; everything else mirrors what I look at every morning.

7/7 services up · 28/28 modules green · 99.97% availability over 90 days · 127/127 deploys with zero rollbacks. Total infrastructure for everything on this page runs at $100.22/mo. The dashboard tile on the home page links straight to the public view.

A spec written by one author, with one model checking it, ships with contradictions and missing requirements that get expensive to fix once code is written. Single-model review is a single point of view. Single point of view is a single point of failure.

Eight phases from intake to traceability. Opus, GPT, Gemini, and Grok all run side-by-side on the same spec. I pick the one that's right for the job, not the one I'm loyal to. The models debate, contradictions are surfaced, missing requirements are flagged. Full debate audit logs persisted alongside the spec so any decision can be traced back to which model said what and why.

Around 12 contradictions resolved and 4 missing requirements surfaced per typical spec, before a line of code gets written. Patterns proven here graduate into client work: TrailLog's tri-model routing, Hopper Ops' synthesis prompts, and the pro bono builds' intake flows all came through Zero Drift first.

A 22-coach youth football league running on a Facebook group, a parent text chain, and three Google Sheets. A Scout troop with parent comms scattered across email, text, and a website nobody updates. Both serving hundreds of families on volunteer time, both hitting the wall where the manual work eats more hours than the program itself.

Real applications, not portfolio exercises. Same scoping discipline, same documentation, same monitoring at Hopper Ops as the paid work. Phone-first where the volunteers actually live. Offline-tolerant where the wifi is unreliable.