"HelixFlow AI team on AI Infra Summit Hackathon Hackathon"
Team Idea
Raw genomic VCF files from DNA sequencing contain hundreds of uninterpreted variants, creating a severe bottleneck where clinicians spend hours manually searching databases to evaluate patient risk. To solve this, we built an auditable, human-in-the-loop decision-support platform that transforms raw genomic data into physician-ready reports without risking AI hallucinations. Our pipeline parses VCF files to isolate high-risk variants, then executes live API queries to MyVariant.info and NCBI PubMed to retrieve real-time annotations and literature abstracts. A Gemini RAG engine synthesizes these findings into structured reports anchored by mandatory inline citations ([ClinVar: ID], [PMID: ID]). Governed by a LangGraph state machine, the system enforces an interrupt state where drafts are held for clinician review and revised via a feedback loop prior to sign-off. The entire workflow is traced via Langfuse for enterprise transparency.
.png&w=256&q=75)
Submission
Team Leader hasn't made a submission yet