redteam
Adversarial agent-pair code-review harness · open source (Apache-2.0)
One model writes the code, an independent model from a different family reviews it adversarially, and a human gates only the irreversible steps. It was extracted from the workflow of building products with AI agents, to stop rubber-stamp review.
GitHub →
Related post →
This blog — Ascendy Engineering
Bilingual (Korean / English) engineering media
Decisions, tradeoffs, and retrospectives from working with AI agents — written with LMO in mind so both human readers and AI answer engines can pick it up and cite it. The subject of this blog is also how this blog is made.
All posts →
AI photo app · multimodal search
Personal project · production stack
AI captioning, tagging, and embeddings for personal photos and video, plus semantic search over complex Korean queries — designed and operated at production grade by one person. The record of the decisions and dead ends in that infra is the AI serving evolution series.
AI serving evolution series →