I'm a solo developer, assisted by AI, focused on long-context LLM competency: local-first memory techniques, portable ways of interacting with models, and making AI plug-&-play — an AI sysadmin capable of wiring into the host board's firmware-level controls. Around that research I build an ever-growing software suite of wired functions, tests, and logs, as the test-bed for bare-metal performance of AI on Linux.

Splinter-Memory - Long-context curation-layer capable of selecting and retaining valuable context within memory before FIFO, while allowing the useless context to be forgotten, by using a cheap CPU encoder. Beats FIFO in all preliminary testing. white-paper

HiveBench - Customisable harness with deep tool capabilities, capable of long-context curation using Strata-Memory with settings for fine-tune, plus the falsifiable evaluation suite Splinter-Memory was built upon. (predictions P1–P11 with measured verdicts).

Note: Unless otherwise stated, the source code for my current projects is hosted in private repositories for intellectual property and security reasons.