Robert Drake

Sole architect and engineer: iOS app, REST backend, content platform, Android port, and the hardware underneath.

• Shipped v1.2 to the App Store. SwiftUI + SwiftData client with an offline-first delta-sync layer, over a Swift Vapor REST API on Fly.io (PostgreSQL, Redis, JWT). ~107,000 lines across 565 Swift files, 524 backend tests, six live subscription tiers.
• Designed the predictive weight engine: a rolling 14-day per-bird burn-rate calculation with an explicit, surfaced fallback when the data window is too thin to trust. The app tells you which mode it is in rather than hiding the uncertainty.
• Built the content platform in Next.js: 50 state-by-state regulatory guides and 100+ reviewed articles, drawing ~3,400 monthly visitors with roughly half from organic search.
• Ran a six-pass pre-launch audit of my own codebase, each finding required to cite file and line. Closed 8 criticals before the first paying customer, including a StoreKit receipt-forgery path and a certificate chain I was checking the root of but not the links.
• De-risked hardware before spending on it: a BLE perch scale on ESP32-S3 + TI ADS1232 with parametric OpenSCAD CAD, and a published noise budget establishing 0.013 g resolution at 80 SPS.

Embedded with client organizations to map how the work is actually done, then build and hand off the system that supports it.

• dbloc.org — deployed an invite-only member platform into a ~70-member organization and its board. Next.js 16, Neon PostgreSQL + Drizzle, Better Auth issuing bearer tokens to an iOS companion app, Stripe, Resend, Vercel Blob. Roles authorized server-side, not hidden in the UI.
• Delivered in phases reviewed with the client at every gate, against a written PRD whose requirement IDs are cited in the source itself, so the code and the spec cannot drift apart quietly.
• Built the regression discipline they had no way to buy: Playwright scripts that drive the live app as each real persona, assert against the database, and clean up after themselves, running in CI beside format, lint, typecheck, tests, and build.
• Delivered and handed off production systems for five regional businesses: tree service, lawn care, falconry outfitter, pest-control hardware, and a fishing tournament with live standings.

This site argues that deployment is the hard part, so it deploys something and publishes the score.

• An LLM triage agent in shadow mode on live inbound. It classifies, scores fit, and drafts. It has never had send permission. Of six decisions in the pipeline, exactly two are model judgment.
• The eval report is public, failures included. 14 golden cases across six dimensions, 12 pass, p50 10.1 s, $0.015 per case. The two failures are named on the page: low-signal messages land under the 0.6 confidence gate and route to a human instead of acting.
• Corrections made in the review queue become regression cases and prompt lessons, so a failure fixed once cannot quietly ship again.

Where the method came from. The process document is never what actually happens.

• Legacy Transportation / DME Expo Logistics — trade show freight across transport, storage, and on-site execution; built the tracking and reporting tooling the team ran on.
• Smartstage Ltd — logistics coordinator, and project manager / site foreman on large installations. International freight and customs; Manhattan Active WMS and Yardview; built the reporting automation that replaced manual tracking.
• EnrichedLeads — founder. A B2B enrichment pipeline and an AI qualification agent, closed when the upstream data provider was served a cease-and-desist. A lesson about auditing the dependency you do not control.
• Nevada Department of Wildlife — raptor rehabilitator. Triage, wound management, fracture stabilization, and conditioning for release. The domain expertise FalconryLab is built on.