ANDROID / VEHICLE OPS Live on Play Store
IntelliAuto
Vehicle management. AutoMind AI, ML Kit OCR, offline encrypted storage.
KotlinFirebaseOffline-firstML Kit
SYS.01 Systems engineering · local AI · product
I build systems that stay understandable under pressure.
Enterprise Linux operations, local inference experiments and production Android apps, documented with the trade-offs left in.
Enterprise operations Local AI lab Shipped software
Enterprise Linux
Lifecycle, automation, hardening, HA and the operational work that keeps platforms predictable.
Select a node to inspect the system
01
Production Apps
Useful software, not AI wrappers.
Two privacy-minded Android products built around local-first value and resilient cloud boundaries.
ANDROID / FINANCE OPS Live on Play Store
IntelliFlow
Finance tracker. Offline-first, cloud sync, budget projections, spending summaries.
KotlinFirebasePrivacy-firstForecasting
02 Educational Series
The LLM Infrastructure Handbook
An engineer's guide to Local AI. Demystifying Transformer architectures, KV cache quantization, modern attention and MoE routing.
03
57 View Full Archive Recent Field Notes
Measured, failed, fixed, documented.
Arc Pro B70 clean suite: Gemma 4 31B MTP, MoE prefill, and Grok tools
Real single-stream timings from a 2026-07-16 B70 suite: long-prompt prefill near 1.7k t/s on MoE, dense Gemma 31B +51% decode with MTP-4, and Grok Build CLI with tools enabled.
02 03 04 05 06
Grok Build CLI with local models on llama-server (Arc Pro B70)
local-ai automation
The Math Behind KV Cache Quantization: Why I Stopped Using Q5_0 for Keys
local-ai
The Reality of Edge AI Research: Why TurboQuant on Intel Arc SYCL Failed (For Now)
local-ai
Breaking the 67 tok/s Barrier: Optimizing Intel Arc Pro B70 for High-Concurrency MoE Inference
local-ai
KV Cache Quantization and Context Ceilings on Intel Arc Pro B70 32GB
local-ai infrastructure