Principal AI Engineer
Aric Camarata
Principal AI Engineer with 20+ years building distributed systems, AI infrastructure, and full-stack applications across defense, enterprise, and product companies. I work on multi-model AI routing, hybrid retrieval (RAG), federated GraphQL, and high-throughput Go backends, and I built nSelf, a self-hosted PaaS, and nClaw, a multi-model AI router. I led development at Northrop Grumman under a TS clearance, now expired. Available fully remote, or on-site in the SF Bay Area up to one week a month.
Earlier work spans Apple, AT&T, the U.S. Treasury (GSOC), Credit Karma, Walmart Labs, and Cigna, across consumer platforms, government security operations, and enterprise health tech. I build AI that keeps data private: local inference on Apple Silicon, classifier-driven model routing, and on-premise RAG pipelines that keep sensitive data off third-party clouds. The open-source nSelf CLI stands up a full Postgres, Hasura, and auth stack on a $4 a month Hetzner server, with the capabilities you would otherwise pay a managed host for.
Work Experience
1420+ years of enterprise and startup development. Contract and full-time work spanning web platforms, backend systems, and open-source tooling.
My Projects
1010 open-source packages and platforms published on npm and GitHub.
Blog
View all (11) →Where Retrieval Stops Being Enough for Agent Memory
Similarity search fails agent memory in two specific ways. Here's how Cascade fixes both, including the RRF fusion code and why recency is a retrieval channel instead of a filter.
ReadCascade Today: An All-in-One Layer for Claude Code
Routing, retrieval, and memory around a coding agent. The tier system that picks the cheapest capable model, why the RAG index degrades gracefully, and what a month of building it settled.
ReadCurtain: Closing macOS Screen Sharing's One Privacy Gap in an Afternoon
macOS Screen Sharing leaves one real privacy gap. The fix turns on filtering input by event source, which lets you block the desk without blocking yourself. Ninety minutes to a working version.
ReadLet’s talk
Open to principal, staff, and senior IC roles in AI infrastructure and developer tooling. Fully remote, or on-site in the SF Bay Area up to one week a month.
