about

i design the epistemic architecture of agentic systems — how machines represent knowledge, reason under uncertainty, and adapt through reflection. i translate ontology, epistemology, and hermeneutics into production cognitive systems: memory structures, confluence models, and feedback loops that turn human judgment into structured machine cognition.

three of those systems run in production. the method is the same each time: encoding human expertise into a working system. in two, i encoded a client's expertise. in ghost, it's my own — i'm the sme, the architect, and the engineer.

ghost — a two-layer cognitive trading system on google adk: a gemini ensemble that establishes the day's informational baseline, and a claude code trading desk that updates, challenges, and acts on that view throughout the session. vertex ai + cloud run, ci/cd — ~85% deterministic pipeline, bayesian regime-isolated feedback, information-theoretic surprisal, ~50k+ lines.

lfx-ai — a lead-generation pipeline on agno (aws) for the linux foundation: enrichment → scoring → contact discovery → recommendations across salesforce / snowflake / apollo, 129× faster via composite-key caching. shipped on contract, clean handoff.

article selector — a curation pipeline on adk (gcp) that encodes a security sme's editorial judgment as a rubric: eligibility → scoring → cohort calibration → selection, where each article is re-scored relative to its cohort over repeated passes. powers a linux foundation security newsletter at 100% precision/recall on its labeled set.

the ethos goes back decades — behavioral economics (kahneman, tversky, ariely, taleb): how human judgment works under uncertainty. encoding that judgment into systems is the work. ghost's bayesian feedback loop is that thinking in code.