At Zerikai, I build production AI infrastructure, including zerikai_memory, a local-first memory layer for AI agents. Its hybrid retrieval and evidence-judgment pipeline checks passages for relevance, support, contradiction, and prompt-injection risk, and refuses unsupported queries.
I started zerikai_memory in response to frontier-model price increases and have iterated it across retrieval, judgment, and GPU residency. Structured tool results reduced agent context overhead from approximately 40% of a 128K context window to 1.6%, while local operation keeps marginal API cost at zero.
I also shipped KillShot, an autonomous four-stage lead qualification pipeline, and built Zerikai AIO Scanner, a Django SaaS for auditing website AI-readiness. Previously, at Compagio, I worked on Amazon SP-API financial reporting and multi-tenant protections; at LegalBit, I developed a Django legal CRM and Twilio communications workflows.

