A 6-part engineering series exploring what it really takes to build AI Playground — a private, 100% offline-first, on-device multimodal AI platform for Android and iOS.

Part 1 of 6 10 min read

I Wanted the Model to Stay on the Phone

Why running language models locally matters for privacy, exploring llama.cpp and ARM NEON acceleration, and the realities of moving beyond toy chat boxes.

Offline runtime architecture · llama.cpp · ARM NEON vectorization Read Part 1
Part 2 of 6 9 min read

The Day a Chat Box Became a Document Engine

How a simple 'attach file' button turned into an offline ingestion pipeline for 25+ formats using pure RFC 1951 Deflate and memory-mapped chunks.

Pure-KMP Deflate · 25+ file formats · Zero-allocation chunking Read Part 2
Part 3 of 6 10 min read

The Model Was Fine. The Phone Wasn't.

Handling the hardest bugs: Android Low Memory Killer (LMK) evictions, vendor GPU driver crashes on Vulkan, thermal throttling, and native lifecycle management.

Android LMK · Vulkan driver crash isolation · Watchdog loop Read Part 3
Part 4 of 6 11 min read

I Put an Image Generator Inside the Phone

Integrating on-device diffusion models with Google LiteRT and Bonsai Image Ternary 4B, managing dynamic feature delivery, and preventing mobile GPU overheating.

Google LiteRT · Bonsai Ternary 4B · Dynamic Feature Delivery Read Part 4
Part 5 of 6 12 min read

At Some Point I Stopped Building a Chat App

Refactoring a monolithic mobile app into a multi-module MVI platform with clean offline boundaries, strict state stores, and zero cloud leaks.

Clean Architecture · Multiplatform MVI · Offline Activity Ledger Read Part 5
Part 6 of 6 11 min read

Pushing Mobile Hardware to the Edge

Engineering dual-model speculative drafting for 2x decoding speed, isolating GPU driver segmentation faults in a native fork() sandbox, and streaming fault tolerance.

Speculative Turbo · fork() sandbox · KV Cache Pre-warming Read Part 6