Freelance — currently taking on-device AI projects

Daisuke Majima

Senior iOS / On-Device ML Engineer — Osaka, Japan.

What I do

I take ML/CV/LLM research and ship it on phones, tablets, and AR devices — fast. I maintain CoreML-Models ★ 1.8k+, the de-facto iOS CoreML model zoo. 170+ OSS repos covering YOLO / SAM / LaMa / on-device LLM / ARKit. 10 apps on the App Store under my own developer account. 120+ converted models on Hugging Face, and I'm a member of litert-community (Google LiteRT). 7+ years iOS / on-device ML.

Recent work

2019 — present · Freelance — Pebble Inc., Osaka

On-device AI for iOS and Android — model conversion, latency and memory optimisation, real-time vision, on-device VLM / LLM

Shipped 10+ apps under my own developer account; maintain CoreML-Models and 170+ OSS repos. Currently taking projects.

2024 — 2025 · Ultralytics

YOLOv5 / YOLOv8 mobile deployment team

CoreML / TFLite / ONNX conversion, ANE optimization, mobile reference apps.

2022 — 2024 · Agencia Inc., Tokyo

ML Engineer

Production ML pipelines and on-device inference for camera-based products.

Shipped on the App Store · 10 apps under my own developer account

LiDAR · ARKit

SnapMeasure

Real-world 3D measurement using iPhone Pro / iPad Pro LiDAR. Tap two points → get a distance you can trust.

CoreML · Vision

Models Zoo

On-device gallery of dozens of CoreML models you can run live on your phone — segmentation, detection, pose, style transfer.

Diffusion · On-device

super2x

On-device image upscaler. Runs ESRGAN-style super-resolution networks fully on the Neural Engine — no server.

Generative · Inpaint

Mask2Face

On-device face inpainting / restoration. Repairs missing or masked regions using a CoreML-converted GAN.

Image-to-Video

AnimateU

One-image animation. Drive a still portrait with motion from another video — on-device.

Diffusion · Image-gen

AIPainter

Text-to-image + inpaint on-device. Stable Diffusion variants ported and optimised to run on iPhone.

Depth · Camera

Blur.

Depth-aware bokeh from a single image. Foreground / background separation through on-device monocular depth.

Memo · OCR

Memosh

OCR-powered quick-capture memo. Photograph anything → searchable text on-device, private.

Utility

Tasmon

Visual task manager with iCloud sync.

Photography

timestone

Long-exposure + light-painting style photography on iOS.

See all on the App Store →

Open source · 170+ repos · most-used model zoo for iOS Core ML

★ 1.8k+

CoreML-Models

De-facto iOS Core ML model zoo. Ports of YOLO, SAM, LaMa, Tesseract OCR, on-device LLM bindings, Stable Diffusion variants, segmentation, depth, pose, style transfer.

github.com/john-rocky/CoreML-Models →

120+ models

Hugging Face — mlboydaisuke

Converted on-device models for Apple Core AI / Core ML and LiteRT: Gemma, Qwen, LFM2, Ministral and more. Member of litert-community (Google LiteRT).

huggingface.co/mlboydaisuke →

Leaderboard

DeviceMark

On-device LLM leaderboard for iPhone — reproducible latency and memory numbers across models and runtimes.

devicemark.github.io →

Stack reference

Repos covering

YOLOv5/v8 · SAM · LaMa · Tesseract · MobileNet · DepthAnything · Style-transfer · Pose · OCR · on-device LLM (Llama.cpp / MLX) · Stable-Diffusion · Real-ESRGAN · NeRF · Gaussian-Splat.

github.com/john-rocky?tab=repositories →

Stack

Swift SwiftUI UIKit CoreML ARKit RealityKit visionOS Vision AVFoundation Metal Combine Swift Concurrency MLX llama.cpp PyTorch HuggingFace diffusers coremltools ONNX TFLite Python Go TypeScript Postgres AWS Kotlin NNAPI

Services · packaged on-device ML deployment

📦 iOS On-Device LLM / CV Production Sprint

Fixed-scope, 2-week engagement: ship your model on iPhone, in production.

For AI teams who have a working PyTorch / Hugging Face model but no iOS specialist to land it on Apple Neural Engine at the latency / thermal / memory budget the product needs.

You bring: your trained model (LLM, VLM, CV, ASR, or diffusion) and a target use case.

I deliver:

  • Model conversion — PyTorch / HF → CoreML / MLX / GGUF / ExecuTorch / ONNX Runtime
  • Apple Neural Engine quantization (FP16 / INT8 / 4-bit) and graph hand-tuning
  • iOS SDK or sample-app integration (Swift / SwiftUI)
  • Latency / thermal sustain / memory benchmark report on iPhone 15 Pro & iPad M4
  • TestFlight-ready build + App Store submission notes

Timeline: 2 weeks fixed.

Price: US$15,000 – US$25,000 per engagement (scope-dependent).

Async-first. JST + a few hours of US-Pacific evening / EU morning overlap.

Discovery / feasibility study (single week, US$5,000) available if you'd like a written go / no-go before committing to the full sprint.

Smaller, well-scoped pieces are welcome too — a single model conversion, a latency investigation, a benchmark on your target device. Tell me what you need and I'll quote it.

Good fit

👍 You'll get value from this

  • You have a working model in PyTorch / HF and need it on iPhone or iPad
  • You're an AI startup / lab without an in-house Apple Silicon specialist
  • Latency, thermals, or ANE perf is the thing slowing you down
  • You want benchmark data before committing to a longer engagement

Not a fit

👎 Skip if

  • You only have a research idea, no model yet
  • You want full-app development end-to-end (this is the inference layer, not UI/UX)
  • Android-only, React-Native-only, or Flutter-only deployment target
  • Strict CET / PT / ET overlap requirement

✉ Inquire — rockyshikoku@gmail.com · plain text is fine, tell me what model and what target device.

Get in touch

I reply within a day. Open to a quick intro chat anytime.

✉ Email rockyshikoku@gmail.com