CuraVar
CuraVar is a "show your work" assistant for classifying genetic variants.
Core trick
An LLM proposes cited ACMG evidence rules while deterministic code computes the score, with each step written to a hash-chained ledger.
Description
CuraVar is a "show your work" assistant for classifying genetic variants. Claude reads the evidence and proposes which ACMG/AMP rules apply — each with a citation — then deterministic code does the final scoring, so the AI never touches the number and the same evidence always gives the same answer. Every step is written to a tamper-proof, hash-chained ledger that can be re-verified on demand. Built for the Built with Claude: Life Sciences hackathon. Research decision-support only — the final call stays with a qualified professional.
README excerpt
# CuraVar **An auditable assistant for classifying genetic variants — it does the legwork and shows every step, so a human expert can check the reasoning instead of trusting a black box.** Built for **Built with Claude: Life Sciences** · Development track ### 🔗 [Live demo → curavar-web.vercel.app](https://curavar-web.vercel.app/) - **Engine:** ACMG/AMP 2015 combining rules + 2018/2020 Bayesian points - **Interface:** Web (React + FastAPI) · offline and live modes - **Status:** Working · 24/24 engine tests · 36 API tests · 100% on validation set --- ## The problem Everyone carries thousands of small differences in their DNA. Most are harmless; a few cause serious disease. When a clinical lab finds one in a patient, a specialist has to decide whether it's harmful or harmless — following an official rulebook (the ACMG/AMP guidelines) that weighs many lines of evidence, and defending that call to a review board. It's slow, high-stakes, and doesn't scale to the flood of variants modern sequencing produces. Worse, a large share of results come back "uncertain," and different labs often disagree because they apply the rules inconsistently. ## What CuraVar does You give it a variant. CuraVar then: 1. **Gathers the evidence** — population frequency, clinical-database assertions, computational predictors, gene mechanism. 2. **Uses Claude to propose which official rules apply** — each with a citation to the specific evidence that supports it. 3. **Double-checks that reasoning*