iOS and full-stack engineer, 13 years. Small teams, usually as the person who takes something from architecture all the way to the App Store alone.

For the past ~18 months I've moved my working process to AI coding agents — as the primary implementation path for shipped, paying production software, not as autocomplete.

iOS / Apple platform Swift, SwiftUI, TCA. Tuist multi-module projects, Fastlane for signing and release. Most of my recent depth is in the parts of the platform where the official docs run out:

• Screen Time stack (FamilyControls / ManagedSettings / DeviceActivity) — DeviceActivityMonitor extensions run in a separate process under a hard memory budget and get killed silently; ApplicationToken is opaque and doesn't survive process boundaries cleanly; restrictions written to ManagedSettingsStore are persistent, so "it unblocks itself at midnight" is a permanent-lockout bug waiting to happen unless something you own can actually run the cleanup.
• Apple MDM protocol — checkin/command over XML plist, DEP/ADE zero-touch enrollment (ABM public-key upload, encrypted token decryption), APNs HTTP/2 over persistent connections, certificate-chain automation, and migrating a fleet off Jamf onto a self-hosted server.

AI engineering Custom Claude Code skills, hooks and subagents built around a real Swift/Tuist/Fastlane codebase — module scaffolding, build verification, code-signing diagnostics. Agentic workflows wired end to end: vision → structured extraction → multimodal generation → browser automation → publish. Tool calling, prompt caching, lightweight RAG, and local model selection under real memory/bandwidth limits.

The part that turned out to matter isn't the model. It's knowing which modules are safe to delegate, encoding the conventions nobody ever wrote down, and being able to tell when the output is confidently wrong.

Web React and TypeScript since 2013 — TanStack, Jotai, Relay/GraphQL, Next.js, Tauri for desktop. Deno + Hono + Zod on the backend.

Network engineering at BUPT. That's why protocol-level work — packet capture, WebRTC, native-to-web bridging, Apple's MDM wire format — tends to be where I'm most useful.

Short, well-scoped contract work. Remote, async, written-first.

• Apple system-level work: Screen Time, MDM, device management
• Getting AI coding agents landed in an existing codebase and CI
• Full-stack delivery for small teams

📮 dabuddha@126.com