"Harnessing Apple Silicon M1 16GB ram to build high-performance AI: projects include a voice-enabled chess and RL agent, a custom GPT transformer trained from scratch, and a growing library of experimental designs and builds."
A floating AI orb for macOS. Voice-first, privacy-focused, local LLM + cloud fallback, chess engine, RL Arkanoid agent, persistent memory.
Swift + Python →A GPT-style transformer trained from zero on TinyStories. Custom BPE tokenizer, full training pipeline, chat UI, LoRA fine-tuning.
PyTorch + Flask →Animations, 3D sculpts, game experiments, and one-off builds — the scratchpad where ideas get tested.
Experiments →In 2026, you don't build an LLM — you train one that already knows the world. A complete guide to LoRA fine-tuning, validation, and deploying a custom model in Ollama on Apple Silicon.
MLX + LoRA + Ollama →Complete technical guide to the LiDAR Scanner Xcode project — time-of-flight physics, ARKit scene reconstruction, Swift mesh capture, and USDZ/OBJ/PLY export on iPad Pro M4.
ARKit · RealityKit · SwiftUI →End-to-end loss prevention using UHF RFID — tag technology comparison, hardware catalog, iPad + Windows app architecture, real-time theft detection, and automated quarterly shrinkage reports.
Swift · Flask · LLRP →6-stage plan to capture the SoloDallas Storm pedal as a neural impulse response — recording 27 measurements, training a TCN in PyTorch, exporting to ONNX, and shipping a VST3 via JUCE.
PyTorch · ONNX · JUCE →Engineering deep dive into the TF-IDF Retrieval-Augmented Generation pipeline added to George — replacing a 6,000-char static dump with sub-millisecond targeted context retrieval. 35 tests · 100% accuracy · 0.98ms latency.
Python · Swift · scikit-learn →Complete beginner's guide to choosing, downloading, and fine-tuning a Hugging Face LLM locally — questions to ask, environment setup, LoRA training, data strategies, and real-world use cases.
Python · transformers · LoRA →Full LoRA fine-tuning workflow on Apple MLX — balanced dataset engineering, loss diagnostics, stress testing, model fusion, and a live spam filter that catches phishing attacks without flagging your friends.
MLX · LoRA · Mistral-7B →A complete line-by-line deep dive into running local AI on Apple Silicon — virtual environments, Metal GPU math, tokenization, KV cache, and the full data flow from keypress to response.
MLX · Python · Apple Silicon →Next project in progress...
Coming SoonNext project in progress...
Coming SoonNext project in progress...
Coming SoonNext project in progress...
Coming SoonNext project in progress...
Coming SoonNext project in progress...
Coming Soon