Open-source human genome analyzer

Got it — you need text for THE BIG PROMPT (staged into milestones) and optionally WHY BUILD THIS?. Here's both, ready to paste. THE BIG PROMPT Build an open-source web app that ingests the artifacts from a local whole-genome sequencing pipeline and renders beautiful, interpretable dashboards. It is an exploratory triage tool, not a clinical diagnosis product, and the UI must reflect that framing everywhere. Input files it must parse: fastp.json (QC metrics), samtools flagstat / BAM index stats (mapped/unmapped counts), samtools depth summary (coverage breadth), the VEP-annotated VCF (large — stream, don't fully load), a multi-disease PanelApp findings JSON, Kraken2 report + Bracken abundance TSVs (contamination), and PLINK2 ancestry projection/assignment TSVs. Stage it into fundable milestones: M1 — Ingestion + schema. File parsers for each artifact type, a normalized internal schema, and streaming support for the multi-GB VCF. Validation and graceful handling of missing files. M2 — QC & Alignment dashboard. Read counts, Q30, mean read length, duplication, insert size, nominal coverage, and a GC-content flag (warn against the expected ~40–42% human band). Mapped/unmapped donut and coverage-breadth bars (≥1x/≥10x/≥20x). M3 — Contamination dashboard. Kraken2 composition, Bracken genus/species refinement, and a derived read-level contamination estimate shown clearly as approximate with caveats. M4 — Variant & panel-findings dashboard. VCF rows scanned, panel-matched rows, prioritized findings, per-panel status buckets, and a "Most Relevant Findings" table (gene, ClinVar significance, VEP impact, # panels matched). M5 — Ancestry dashboard. PCA scatter of the sample projected onto reference panels (1000 Genomes, HGDP650) plus a consensus card — framed as a reference-relative match, never a % composition or identity claim. M6 — Polish + safety framing. Top-row status cards, pipeline-order navigation, per-section "what this means" captions, and enforced data-quality banners. Hard UX rules to enforce throughout: never show disease likelihood, risk %, or prediction language — use "finding for review"; label the contamination estimate as read-level and approximate; present ancestry as a PCA reference match only; and surface GC-content and low ≥20x coverage as data-quality warnings that gate interpretation. Stack: React frontend, streaming parsers for large files, all processing local (no uploading raw genomic data to third-party services). Ship open-source under MIT.