Core ML On-Device Inference: The Patterns That Actually Ship
Core ML runs models on Neural Engine, GPU, or CPU. The patterns that ship: model conversion, dispatch hinting, latency budgets, and quantization.
AI & TechnologyThoughts on design, development, AI infrastructure, and building products.
Core ML runs models on Neural Engine, GPU, or CPU. The patterns that ship: model conversion, dispatch hinting, latency budgets, and quantization.
AI & TechnologySwift Testing ships with Xcode 16+ and Swift 6. Macros, traits, parametrized tests, parallel by default. Migrate incrementally; UI tests stay in XCTest.
AI & TechnologyEvery iOS app capability faces three surfaces: human, Apple Intelligence, agent. Each has different obligations, rendering, latency, and trust posture.
AI & TechnologyThree callers can write to the same shopping list: a human, Apple Intelligence, and an external agent. Truth has to live somewhere. Pick the substrate.
AI & TechnologyAn iOS agent has the developer's signing keys, builds, and project file reachable. Hooks bound the blast radius. Four shipping patterns.
AI & TechnologyTwo LLMs touch a Swift app. The on-device model that ships with the app and the agent that wrote the code. Different stacks, different obligations.
AI & TechnologyA blog cluster's voice comes from what it refuses to publish, not what it ships. Categorical, pattern, and interesting refusals each shape what the cluster is.
AI & TechnologyiOS 26's Foundation Models framework puts a 3B-parameter LLM on every Apple Intelligence device. The Tool protocol is the surface that makes the model useful.
AI & TechnologyReturn's Live Activity looks like a Lock Screen countdown. It is a five-state lifecycle machine with three dismissal paths. The production code and bugs.
AI & TechnologyTwo protocols, one app. App Intents expose your app to Apple Intelligence. MCP exposes the same domain to Claude, ChatGPT, and the rest. The routing question.
AI & TechnologySwiftData's API is two macros. The cost is what happens after you ship. Optional fields are the cheap migration; non-optional adds need a VersionedSchema.
AI & TechnologySwiftUI is a result-builder DSL on top of a value-typed View tree. Once the substrate is visible, AnyView, Group, and ViewBuilder stop being mysterious.
AI & TechnologyTechnical writing published at Introl
Comprehensive hardware recommendations and cost analysis for running large language models locally.
GPU selection guide comparing NVIDIA's latest datacenter accelerators for different AI workloads.
Deep technical dive into Google's Tensor Processing Unit evolution from TPUv1 to TPUv5.
Resource sharing strategies for GPU clusters in containerized environments.
Guide to building and managing distributed AI computing with Ray framework.
Analysis of open source LLM economics and DeepSeek's competitive positioning.
Future datacenter power requirements and NVIDIA's next-generation GPU roadmap.
Small modular reactor solutions for powering next-generation AI infrastructure.
Technical analysis of DeepSeek's Multi-Head Compression architecture innovations.