Portland-based developer. Currently obsessed with local-first AI and reclaiming the reasoning power of open weights.I recently published "UNSHACKLED," a technical manual for running unrestricted 405B models on consumer/prosumer hardware. My goal is to bridge the gap between "censored cloud APIs" and "sovereign local inference."
Working on: Testing orthogonalization techniques (abliteration) to remove refusal layers in Llama 3/4 models without destroying their core reasoning capabilities.
Site: https://www.aimanual2026.com/unshackled X/Twitter: Unshackled_AI Email: richstuart@gmail.com