{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/4e9ab966e3c848828d0ebbe9b33c7b47\" frameborder=\"0\" width=\"1670\" height=\"1252\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":1252,"width":1670,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":1252,"thumbnail_width":1670,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/4e9ab966e3c848828d0ebbe9b33c7b47-4bfecf7d832629a2.gif","duration":185.317,"title":"Louis: Trust-Graded Autoimmune Drug Target Discovery","description":"🔗 Try it live: https://claude.ai/code/artifact/b4a412b5-2eeb-4447-992c-0771fad5398c\n💻 Code (open source): https://github.com/rpinho/louis\n\nLouis is a drug-target discovery assistant for autoimmune disease that grades every answer by trust and learns from lab results. It runs on a genome-scale CRISPRi screen of human CD4⁺ T cells — the Marson–Pritchard autoimmune map — which no lab can use raw: you need a bioinformatician, and you can't tell if an answer is trustworthy. Louis fixes that. It returns graded, checkable evidence and flags non-peer-reviewed signal — it even stress-tests its own flagship pick, taking DOT1L from grade A down to C. It designs the next experiment (a two-arm go/no-go knockdown) and updates grades when bench results come back, as with HDAC7 for Th17-driven colitis. And it reads the bleeding edge: off-allowlist signal about a target like DOCK2 — from X, Bluesky, and conference abstracts — surfaced as signal, never as evidence.\n\nBuilt with Claude Code, Claude Opus, and Claude Science for the Built with Claude: Life Sciences hackathon (Builder track)."}