Software at Obedience Corp. I work on the parts of the stack that have to stay on the machine: local inference, voice pipelines, and the festival tooling we use to keep agent work legible.
Most days that looks like Go, Python, just, and too much time staring at first-audio latency. I like models you can stat, configs you can diff, and CLIs that print the same thing twice.
- Local TTS / STT (Qwen, Whisper, the usual messy device graph)
- camp and fest — how we organize long-running agent work
- Native workers, Metal, and why the second request is always the interesting one
- Company: obediencecorp.com
- Org: Obedience-Corp

