Gregory Kennedy
Lead AI Systems Architect — Keraunos Deterministic AI Agent System
Los Angeles, California, United States · 1 connectionNot on Unlinked · Imported profile. Is this you? Claim it
Is this about you? Request removal
About
The AI industry generally ships non-deterministic "hope for the best" vibe coded solutions. I build: deterministic, auditable, human-in-the-loop AI pipelines that will not silently ship bad output.For the last four years I've worked at the intersection of three things the field urgently needs more of:1. AI agent harness engineering. I designed and built Keraunos, a propriatary 10-stage deterministic AI agent pipeline: alternating hard-coded verification nodes (linting, security scanning, CI enforcement) with bounded LLM agent loops, backed by a 64-tool MCP toolshed. The core principle: the system runs the model, not the reverse. If the model can't fix an issue in two rounds, it stops and escalates to a human. An AGENTS.md passive-context strategy took agent eval pass rates from a 53% baseline to 100% against Vercel's hardened Next.js 16 evals. Keraunos now ships two production verticals: audit-ready technical training courseware and jurisdiction-verified legal documents, protected by a fail-closed Citation Verification Gate that achieved zero false acceptances against an adversarial corpus including the actual fabricated authorities from Mata v. Avianca. 2. LLM fine-tuning and dist
Experience
Feb 2025
Jan 2025
Aug 2022
Mar 2018 to Dec 2021
Jan 2024 to Mar 2026
Jun 2025 to Aug 2025
Jun 2021 to Dec 2022
Jun 2017 to Apr 2020
Jun 2018 to Nov 2019
Jun 2012 to Jun 2017
Jun 2016 to Jun 2016
Jan 2003 to Dec 2012
Sep 2010 to Jun 2014
Feb 2012 to Jun 2012
Mar 2011 to Jun 2011
May 2009 to Jun 2012
Mar 1998 to Jun 2002
Education
Bachelor of Science - BS Computer Science
1998 to 2003