Team

A woman with shoulder-length blonde hair, wearing a black top and silver jewelry, looking directly at the camera against a plain white background.

Abigail Spach

Abigail Spach is the founder and CEO of Luminae and SouthHaus, Inc., building an inference-time AI audit layer for defense and mission-critical enterprise systems. She started Luminae from a simple conviction: AI decisions cannot be taken on faith — they have to be seen, evaluated, and proven in real time. DARPA candidate for KDMA PPDMA artificial autonomous systems.

A systems engineer with a decade of experience in UI/UX and large-scale systems architecture, Abigail led the digital transformation work for the Italian Trade Agency, rebuilding national export and onboarding systems used by thousands of manufacturers entering the U.S. market. Her background spans global commerce, government programs, and AI infrastructure, with a focus on how humans and machines actually make decisions under pressure.

At Luminae, she is architecting the runtime layer that sits alongside models and agents, inspects every decision in milliseconds, and emits signed, replayable evidence for operators, risk teams, and defense partners. Her work bridges defense and enterprise, with an emphasis on AI accuracy, behavioral drift, and long-horizon responsibility as AI moves toward AGI and beyond.

A close-up portrait of a young man with short, black hair, slightly wavy, and fair skin, wearing a black hoodie, looking directly into the camera against a blurred light-colored background.

Liu Zhiqiang

Liu Zhiqiang leads the architecture behind Luminae’s inference-time audit layer, turning raw AI decisions into signed, replayable evidence for defense and enterprise systems. A graduate of Tsinghua University, Zhiqiang brings a decade of AI infrastructure experience across global defense and mission-critical enterprise environments, building systems where failure is not an option. YC 23 S

At Luminae, they own the end-to-end technical stack: high-throughput decision pipelines, real-time scoring, cryptographic signing, and multi-tenant isolation that allows any model or agent framework to be audited inline, without slowing it down.