AI Systems · Agentic Software Engineering · Local LLMs · Embedded Systems

I build local AI infrastructure, developer tools and embedded systems. Most of my current work is around local LLMs, agentic coding workflows and connecting software agents to physical devices.

LLM infrastructure · AI agents · developer tooling · ESP32 / IoT · robotics

Local-first AI workspace and control center.

LM-Nexus is my main project. It brings local model management, chat, agents, terminals, files, knowledge, monitoring and system tooling into one browser-based environment.

Highlights:

• managed local llama.cpp / llama-server inference
• GGUF model discovery and management
• OpenAI-compatible local APIs
• persistent multi-provider chat
• custom Agent Runtime with tools, skills and delegation
• approval-gated agent actions
• browser PTY / Windows ConPTY terminals
• file editing and knowledge management
• CPU / RAM / GPU / VRAM monitoring
• Docker and shell-based integrations
• modular extension and package system
• command-line access to models, chat and modules

Python · FastAPI · Vue 3 · TypeScript · Vite · Docker · llama.cpp

Currently private and under active development.

I spend a lot of my development time on agentic software engineering: orchestration, delegation, reusable skills, configuration layering, execution semantics, verification and long-running coding workflows.

I'm an active upstream contributor to oh-my-opencode-slim, an OpenCode multi-agent orchestration framework. My work there covers agent runtime behavior, configuration and Agent Skills, the native Companion UI, testing and CI.

I use my public fork as a contribution workspace for branches and PRs going back upstream, not as a separate distribution.

Selected upstream work:

• Local inference scheduling - #1179: same-provider background tasks can opt into foreground execution to avoid unnecessary model/KV-context switching on constrained local inference backends.
• Agent Skills configuration - #1190, #1204: composable per-agent skill additions/removals and automatic project-local skill discovery, with layered config semantics and path-boundary checks.
• Native Companion controls - #1419, #1421: scoped Project / Global / Inherit preset switching plus live agent/model and waiting-input state.
• Companion CI - #1424: pull-request validation for the Rust Companion on Linux and Windows.

Most of this work starts with tracing existing runtime behavior and ends with regression coverage, cross-platform validation and review fixes.

I keep reusable Agent Skills for OpenCode and compatible coding environments in Aveer/skills.

It includes workflows for:

• multi-agent deep work
• architecture and codebase exploration
• systematic debugging
• verification planning
• repository governance
• worktrees
• TypeScript / Vue development
• product and UX reasoning

I'm building the hardware side of the Nexus ecosystem around a shared ESP32 firmware platform.

Current capabilities include:

• distributed device discovery
• capability-based device APIs
• authenticated device-to-device communication
• HMAC-SHA256 fabric authentication
• capability routing
• signed OTA updates
• runtime hardware configuration
• Zigbee coordination
• mmWave presence radar
• servo control
• OLED, distance and temperature sensors
• embedded dashboards and JSON APIs

The goal is to give AI agents a controlled way to interact with real-world devices and, eventually, robotics hardware.

C · ESP-IDF · ESP32 · ESP32-C5 · Wi-Fi · Zigbee · mDNS

Currently private and under active development.

Small desktop utility for managing oh-my-opencode-slim presets across multiple projects.

It supports separate Global / Project / Inherit semantics, bulk project changes, OMO Slim precedence rules and JSONC-preserving atomic writes without modifying OpenCode itself.

Python · Tkinter

OpenCode setup, validation and asset-generation toolkit.

It can bootstrap project configuration, generate agents/skills/commands, detect config drift, validate CI state and perform safe upgrades with backups.

npm install -g openzeus

Local Korean vocabulary extraction and study application.

Features include morphological analysis, TOPIK-aware filtering, an offline Korean-English dictionary, SQLite study data, lightweight spaced review and Anki-compatible export.

Python · FastAPI · Vue · SQLite

A dependency-free PowerShell utility that improves The Settlers 7 performance by adjusting process priority and CPU affinity to avoid logical sibling threads.

The current maintained implementation is PowerShell-first; the old Python-packaged executable is retained only as a legacy option.

AI / Agents
Local LLMs · llama.cpp · Agent orchestration · Tool calling · Agent Skills · OpenCode

Backend
Python · FastAPI · Pydantic · pytest · SQLite

Frontend
Vue 3 · TypeScript · JavaScript · Vite · Pinia

Systems
Docker · PowerShell · Linux / WSL · Windows · GitHub Actions

Embedded
C · ESP-IDF · ESP32 · Wi-Fi · Zigbee · Sensors · Robotics

Local AI infrastructure · agentic coding · multi-agent systems · developer tooling · embedded systems · robotics · AI-to-device interfaces