Juyoung Kim

Software / Product Engineer · San Francisco, CA · U.S. Citizen

juyoungk23@gmail.com · GitHub · LinkedIn

Download PDF

Skills

Apple platforms: Swift 6 / SwiftUI, RealityKit, visionOS, EventKit, HealthKit, WatchConnectivity, MultipeerConnectivity, SharePlay, App Clips, Swift Testing

AI & backend: LLM integration (Claude, Gemini), computer vision (PyTorch), 3D Gaussian Splatting, Python, GCP, Supabase (Postgres), Next.js

Work Experience

San Francisco, CA
  • Third engineer on Poppy, a proactive AI assistant for iOS, macOS, and watchOS; joined pre-launch in April and shipped the May 2026 public debut (TechCrunch, 9to5Mac). Backed by Kindred Ventures.
  • Co-built Pods, Poppy's long-running agentic tasks that handle goals over days or weeks rather than a single prompt; own the Pods backend and the SwiftUI client that renders agent output as native charts, chips, and deep links.
  • Built cross-device handoff: a suggestion on iPhone or Watch opens the meeting link on the Mac, via a WatchConnectivity relay into MultipeerConnectivity.
  • Built two-way sync with Apple Reminders via EventKit, Poppy's backend as source of truth: offline edits and deletions reconcile on re-sync, and delivery-ownership logic guarantees exactly one alarm fires.
  • Built the health layer merging HealthKit and Oura Ring data (deduplicating overlapping Watch and ring signals) and shipped sleep briefings that explain scores using calendar and message context from the user's actual day.
  • Own product analytics end to end: built the Mixpanel pipeline from zero, dashboards, and weekly metrics reports; shipped a redesigned onboarding flow that reached ~60% completion through OTP verification and paywall.
San Francisco, CA
  • Backend infrastructure on data migration to Google Cloud. Designed and ran the pipeline moving hundreds of millions of records from legacy on-prem systems to GCP SQL with zero data loss, and refactored the Python ETL layer to cut processing time by 60%.
  • Led monthly production deployments integrating changes from three teams under bank change-control.

Independent Products

iOS, Web, visionOS
  • Built a cross-platform app (iOS, web, visionOS) converting a single 2D photo into a 3D Gaussian Splat via an event-driven serverless GPU pipeline; a <15MB iOS App Clip lets recipients view shared memories from an iMessage link with no install. Live on the App Store.
SwiftUI, iOS, visionOS
  • Launched a subscription immersive streaming app on iOS and Apple Vision Pro; signed broadcast partnerships with three U.S. MMA promotions.
  • Built synchronized co-watching with Spatial Personas: a custom RealityKit rendering architecture that keeps friends' Personas seated side by side during stereoscopic video, including live HLS streams, which Apple's native player still can't do; open-sourced it as ImmersiveWatchParty (github.com/juyoungk23/ImmersiveWatchParty-SDK).
Fight analytics from broadcast video (Python, PyTorch, Next.js)
  • Built an end-to-end computer-vision product that turns single-camera MMA broadcasts into verified fight analytics, including strike detection (~96% recall with human-in-the-loop verification), fighter tendencies, patterns, and performance metrics. Shipped a live product now used by MMA promotions across the U.S.

Education & Certificates

  • B.S. in Computer Science, UNC Chapel Hill (2023)
  • Google Cloud Professional Cloud Architect | Google Cloud Associate Cloud Engineer