Applied AI Engineer — local-first agent infrastructure, voice/HCI automation, and developer tooling.

I build local-first AI agents that safely operate machines. My work sits at the intersection of agent runtimes, multi-provider model routing, desktop accessibility, and human-computer interaction — all designed to run on the user's own hardware with their own API keys.

• Agent infrastructure: multi-provider routing, MCP tool protocol, memory, subagents, safety/approval layers, headless delegation
• Voice & HCI automation: push-to-talk STT, TTS, on-screen keyboard, controller-driven desktop control, HUD overlays
• Systems/integration focus: Linux services, systemd, browser automation, local model inference, cloud fallback strategies

I'm self-taught — HVAC engineer by background, systems builder by practice. I build tools that solve real gaps: offline-capable agents, accessible desktop input, and infrastructure that doesn't require a subscription.

A local-first AI agent runtime with multi-provider routing, MCP tooling, browser automation, voice/vision, subagents, and a safety layer.

• ~43K lines Python, 57 test files, terminal + browser extension surfaces
• Routes between Ollama (local) and OpenRouter / Groq / Gemini / Anthropic / OpenAI (cloud)
• Headless delegation mode for CI and agent orchestration
• Approval queue, typed action dispatch, and privacy-cloud guard

Voice-first desktop accessibility for Linux. Control the entire desktop with an Xbox controller — no keyboard, no mouse required.

• Push-to-talk speech-to-text via Groq Whisper
• GTK floating on-screen keyboard and controller legend HUD
• FastAPI voice bridge for STT/TTS on :8002
• Auto profile switching and systemd-managed services

A FastAPI proxy that routes Claude Code CLI calls to local Ollama first, with selective cloud escalation.

• Off-grid mode prevents cloud calls unless explicitly enabled
• Charter injection, persistent task state, permission-mode sync
• Multi-provider cloud tiers: Anthropic, OpenRouter, Groq, Cerebras, Fireworks

An open, HSDS-shaped REST API + assistant for 2-1-1 health and human services data across the US.

• Federates ~200 local/state services behind one unified API
• Assistant actions: summarize, read aloud, print, email, pinpoint documents, fill forms
• Built as a tool for AI agents rather than a standalone app

An MCP server + HTTP bridge for managing client intake and form completion across reentry programs.

• JSON-based profiles and templates — no database required
• Browser autofill handoff payload
• Optional REST bridge for web UI integration

• Claude 101 — Anthropic (May 2026)
• Claude Code 101 — Anthropic (May 2026)
• Introduction to Claude Cowork — Anthropic (May 2026)
• AI Fluency: Framework & Foundations — Anthropic, co-badged UCC / Ringling College of Art + Design / HEA / National Forum (May 2026)

Full records: ebey317/cert-path

• Indianapolis, Indiana
• ebey317@gmail.com
• Portfolio: ebey317.github.io
• GitHub: @ebey317