Daisuke Majima

Senior iOS · On-Device ML Engineer  ·  ex-Ultralytics  ·  🇯🇵 Japan  ·  Remote

Shipping LLMs and CV models that run fully on-device — iPhone · iPad · Apple Silicon Mac.
No server round-trip  ·  No per-token cost  ·  Privacy by default.

 

---

📦 Featured

🧠 CoreML-Models — 1.7k★ The de-facto iOS Core ML model zoo. YOLO · SAM · LaMa · OCR · LLM · Stable Diffusion · depth · pose · style transfer.	🥽 RealityKit-Sampler — 176★ Runnable Apple RealityKit samples: anchors · physics · lighting · gestures · occlusion · mesh · audio.
🖼 SemanticImage — 156★ iOS image / video filter & ML utility — Vision + CoreML + AVFoundation, production-tested.	💬 CoreML-LLM — 130★ LLMs on Apple devices: CoreML · MLX · llama.cpp with quantization recipes.
🧽 lama-cleaner-iOS — 81★ iPhone / iPad on-device LaMa image-inpainting cleanup tool.	🎯 CoreML-YOLOv5 — 67★ YOLOv5 ported to iOS, with Apple Neural Engine notes.

_{+128 more public repos · browse all}

---

🚢 What I do

Take AI models from "works in a notebook" to "ships as a real iOS app."

• CoreML conversion + ANE optimization — PyTorch / HF / ONNX → CoreML / MLX / GGUF / ExecuTorch. FP16 · INT8 · 4-bit. Graph hand-tuning. KV-cache packing.
• ARKit + LiDAR spatial pipelines — depth fusion · scene reconstruction · plane / object anchoring · RoomPlan · Vision overlay · Metal post-processing.
• Production iOS apps — SwiftUI · UIKit · Combine · Swift Concurrency. 30 FPS LiDAR + on-device YOLO + on-device LLM. Sub-200ms recognition on iPhone 15 Pro.

---

🌐 Upstream contributions

---

📱 10 apps on the App Store

_{under my own developer account · apps.apple.com/developer/daisuke-majima}

SnapMeasure · Models Zoo · super2x · Mask2Face · AnimateU · AIPainter · Blur. · Memosh · Tasmon · timestone

---

🛠 Stack

---

📊 GitHub

---

✉ Contact

📧 rockyshikoku@gmail.com  ·  🌐 john-rocky.github.io  ·  💼 LinkedIn  ·  𝕏 @JackdeS11  ·  ✍ Medium

_{Async-first · Plain text fine · Reply within a day}

---

_{Currently solo-lead on production iOS apps combining ARKit + LiDAR + YOLOv8 + on-device OCR + on-device voice + multi-device backend sync at iPad scale. Detailed case studies under NDA.}