01 / RESEARCH 02 / BUILDS 03 / SECURITY 04 / SYSTEMS LAB 05 / RESEARCH WORKBENCH ↗ 06 / SYSTEMS COCKPIT ↗ 07 / FIELD NOTES 08 / ABOUT & METHODOLOGY 09 / DESIGN SYSTEM → PORTFOLIO (HARRYDEV.ONE)

HOME / TAXONOMY

AI & Local Models

On-device quantized LLMs, agent tools, and autonomous pipelines.

PUBLICATIONS (4)

Mobile GPU LLM Acceleration: Vulkan & OpenCL Compute Pipelines on Edge Silicon

Bypassing CPU bottlenecks on mobile SoCs by binding quantized GGUF weights directly to Adreno and Mali shader cores via Vulkan SPIR-V kernels.

ADB MCP Server: Bridging Model Context Protocol to Real Android Hardware

Real-time JSON-RPC 2.0 tool interface over Unix sockets and ADB daemon for autonomous device interaction, visual grounding, and UI exploration.

Building a Local AI Agent That Controls Android

Architecture, ADB transport, local quantized GGUF inference, tool execution and failure analysis on physical mobile hardware.

Building an MCP Server for ESP32 Microcontrollers

Hardware-in-the-loop Model Context Protocol (MCP) server & ESP32 C++ firmware for AI-to-microcontroller bridging over USB-OTG and BLE.