Hi, I'm Jered Leisey, a full-stack automation developer. I build software that takes repetitive work off people's desks.
Check out my latest work and life.
- Dispatch · Underworld · Part 4Eleven Claimed, Two VerifiedThree local agents, one 9B model, a job with 21 milestones, and 150 minutes. At minute 50 the agents claimed 11 milestones and an outside check passed 2. The record of a swarm run where the method worked and the product did not.
- Dispatch · Underworld · Part 3Three Agents and an InboxTwo runs of a three-agent swarm in one day. First in Claude Code on a cloud model, with direct messages between subagents. Then fully local: three sandbox VMs, a 9B model on my own GPU, and an inbox service that sets the sender, writes the log, and owns the files.
- Dispatch · Underworld · Part 2A Voice Assistant That Could Not See ColorA fully local voice pipeline for Home Assistant: Wyoming Whisper and Piper, an LLM conversation agent on Ollama, measured on a GTX 980 and an RTX 4080. Then an audit of the tool calls that shows why the assistant claimed a color change it never made.
- Dispatch · Underworld · Part 1Waking a Server After Two YearsA Proxmox VE 8.2 host that sat dark for two years, upgraded to 9.2 in one day with an AI agent on SSH. A stale ZFS label that stopped the boot, a gateway that was not a gateway, SVM off in the BIOS, and GPU passthrough for a GTX 980.
- ProjectOpen →Jev prompt testerTry prompts against the Jev decision model from TypeSafe AI, with your own questions and criteria.