Full-stack engineer doing AI Ops at BuildPass. I build autonomous agents that do real jobs, like investigating support tickets, drafting CS replies and maintaining the company wiki. I also build the platforms that run them, so a agents can work unattended while staying observable and auditable.

Mostly TypeScript, Next.js, shadcn/ui, Neon, Vercel, Fly.io, and a lot of Claude.

Currently deep in agent harnesses: Claude Agent SDK, Vercel AI SDK, MCP servers, and lately Mastra.

AI agent that reads architecture drawings. Extracts structured data from IFC models and PDF construction documents, cross-validates between sources, and helps builders with quantity takeoffs. Claude Code plugin with IfcOpenShell and pdfplumber CLI tools

AI-powered IFC design toolkit with browser-based viewer. CLI tools create and query IFC building models from natural language, rendered in a 3D viewer with spatial tree, property inspector, and visibility toggles. Live demo

Voice-first AI agent for macOS that can run your computer. Floating liquid glass orb with voice-first interaction, shell and file access, screen capture, and AppleScript automation. Pure Swift, single WebSocket to Deepgram Voice Agent API, BYO any LLM via OpenRouter

Voice-first AI companion for macOS. Floating presence widget, natural realtime conversation, persistent memory, and local computer tools. Python backend with OpenAI Agents SDK, Swift frontend with AVAudioEngine

Voice-first AI agent with sandboxed tool execution. Talk to an LLM that runs shell commands in an isolated Firecracker microVM via Sprites. LiveKit WebRTC transport with Deepgram STT/TTS and any model via OpenRouter

AI employee with its own computer. Persistent cloud VMs run a TypeScript coding agent per user with multi-provider LLM support via OpenRouter. Workspace grid UI presents files, documents, and results as draggable cards instead of chat messages

Document intelligence platform where each user gets a dedicated VM running a Claude-powered agent with a memory daemon that learns across sessions. Agent renders structured extraction results to a streamable canvas

Local AI agent harness with a custom inference loop, tool execution, and streaming chat UI. Runs on-device with Ollama models

AI-powered document extraction platform that turns unstructured documents into structured data. Autonomous agents extract fields into JSONB with confidence scores, support natural language corrections, and batch-extract across stacks

Talk to Revit. Watch it build. AI agent that turns plain English into executed Autodesk Revit commands with automatic error recovery

NCC compliance, without leaving Revit. A dockable side panel for Autodesk Revit 2025 that answers National Construction Code questions using RAG with specific section references

GSC insights dashboard that visualises Google Search Console data and generates AI-powered SEO analysis using Claude. Built as a timed coding challenge.

Local text-to-speech for macOS. Copy text in any app, then use Play Copied Text from the menubar or floating widget. Runs entirely on-device and ships as a native .dmg installer. Download latest DMG

Agent orchestration plugin for Claude Code with multi-session persistence, structured exploration, and multi-agent execution via the Ralph loop

Inter-session messaging for Claude Code. HTTP broker lets multiple Claude sessions discover each other, send direct messages, broadcast, and coordinate via pub/sub channels. Built for multi-agent workflows where agents need to talk across terminal windows

Self-improving harness and autonomous task-learning proof of concept. Stilts combines iterative harness optimisation with a task-bounded autonomy loop for planning, memory, safety gates, and progress metrics.

Personal AI operating system for Claude Code with persistent memory across sessions, subsystem apps (calendar, second brain, networking, work PA), and a hybrid local search engine built on SQLite + sqlite-vec (vector similarity) + FTS5 (BM25 keyword)

Turn-based strategy game where two LLM agents fight each other in real time. Pick any Claude, GPT, or GLM model for each side and watch them go to war. Play here

Distributed IoT system connecting ESP32 microcontroller with multiple sensors and web dashboard via RESTful API and MQTT

Animal shelter management system with CRUD operations, real-time RFID tracking, and team collaboration via Agile practices

RAG document processing pipeline for chunking and vectorizing documents with drag-and-drop upload interface

AI-powered task manager that uses OpenAI API to automatically enhance and restructure user task inputs

Model Context Protocol server enabling AI agents to query and interact with PostgREST APIs

• Building autonomous AI agents that work persistently across sessions
• Designing real-time systems with WebSockets and per-user compute
• Creating developer tools that improve AI-assisted workflows
• Deepening software architecture skills and contributing to professional codebases

• LinkedIn
• fraserbrown@live.com
• Melbourne, Australia