Infrastructure engineer. I build AI agents that operate real systems (telephony, monitoring, configuration management) and keep a human in control of what they do.
- AI on the Line: an LLM that holds real phone conversations. You describe the errand, the model finds the numbers, briefs each call and places it through MCP tools. A voice agent holds the conversation over Asterisk (ARI and AudioSocket) with local Whisper and Piper, voice activity detection, barge-in and keypad tones, then reports back with answers and transcripts. It also screens my incoming calls and keeps salespeople talking. It says it is an automated assistant and never commits to anything on your behalf. One container, live status page.
- Agent-assisted incident investigation. MCP servers that connect an LLM to log search, monitoring, virtualisation and configuration management, so alerts, logs and config changes are correlated in minutes. Read-only by default; changes go through review.
- Retrieval over documentation. Wiki content extracted, indexed and queried through a locally hosted model.
| Repository | What it is |
|---|---|
| ai-on-the-line | Give it a goal and it phones the place, gets past the menu and the hold music, talks to whoever answers and reports back; it also answers my calls and keeps salespeople talking |
| esphome-tesla-ble-multi | Local Bluetooth control of several Teslas from one ESP32 in Home Assistant, set up with just the VIN, with verified replies from the car |
| wifi-association-presence | Who is home and in which room, from the access points you already have |
| ha-pvoutput-uploader | Home Assistant readings to PVOutput with local recording and upload recovery; in the HACS default list |
| wattledger | ESPHome energy monitor as an autonomous meter of record (early development) |
| tesla-ble-fleet | One Home Assistant device per car across several ESP32s (design stage) |
| json-graylog-tcp-logger | Small C process that pipes Apache logs to Graylog over TCP without ever blocking the web server |
Long experience running production infrastructure: Linux, ZFS, MySQL replication, Puppet, network configuration management, Java application servers. Electronics and embedded systems since childhood, which is why much of the above ends at a device.
Open an issue on the relevant repository.



