Blog

Insights & Engineering

Technical articles, industry perspectives, and lessons from the field.

Oct 14, 20263 min read

GPT-6 Reaches Every ChatGPT User — Read the Fine Print

OpenAI is rolling GPT-6 out to free and paid ChatGPT users. Why its candid deployment notes matter, and why new models are never drop-in upgrades for products built on them.

Sep 16, 20263 min read

iPhone Duo and iOS 27: What Apple’s September Event Means for App Makers

Apple’s first foldable puts two screen sizes in one phone. What to check in your app before it ships, plus photo authenticity and the cost of AI showing up as Siri limits.

Aug 19, 20264 min read

Keeping a Local LLM Honest: Guardrails for Gongje’s Corrections

A language model that fixes homophones will happily rewrite everything else. The prompt, output checks and edit-distance limits that keep Gongje’s corrections to the words you actually said.

Jul 15, 20264 min read

Gongje on the Mac App Store: What the Sandbox Changed

One codebase, two builds. How we put Gongje inside the App Sandbox, what we compiled out, and why the App Store version made the GitHub version better too.

Jun 17, 20264 min read

Converting Cantonese Whisper Models to CoreML

How we took community Cantonese fine-tunes of Whisper and converted them to run on Apple’s Neural Engine for Gongje — with the numbers from the conversion log and the models that didn’t make it.

May 21, 20263 min read

Google’s May: Android 17, Gemini 3.5 and Tools Built for Agents

The Android Show and Google I/O 2026 in one read: what was announced, and why agents became a platform audience — from WebMCP to an Android CLI built for AI.

Apr 8, 20263 min read

Gemma 4: Open Weights, Apache 2.0, and Small Models Worth Shipping

Google’s Gemma 4 family spans phones to workstations, and drops its custom license for Apache 2.0. Why the license and the small E2B/E4B models are the real story.

Mar 11, 20263 min read

GPT-5.4: A Million Tokens, and the Quieter Upgrade

OpenAI’s GPT-5.4 brings a 1M-token context window to the API. Why that is headroom rather than a retrieval strategy — and why Tool Search may matter more for agents.

Feb 20, 20267 min read

On-Device AI: Why We Bet on Local Inference

Cloud AI isn’t always the answer. We explore when on-device inference wins — from privacy and latency to offline capability — and the frameworks making it practical today.

Feb 10, 20266 min read

Building Gongje: Lessons from Our First Open Source Project

Behind the scenes of building a Cantonese speech-to-text macOS app with WhisperKit and MLX — why we went open source and what we learned along the way.

Ready to build something great?

Let's discuss your next project and explore how we can bring your vision to life.

Start a Conversation