- contact@insightvariant.com
Multi-agent AI platform that classifies genomic variants using ACMG/AMP guidelines. Automatically mines PubMed literature, queries ClinVar, gnomAD, and OMIM, and generates clinical-grade evidence reports for every variant.
A managed pipeline that takes a VCF, normalises against GRCh37/GRCh38 with HGVS notation, and returns ACMG/AMP-graded variant reports.
VCF + HPO phenotype terms. GRCh37 / GRCh38 auto-detected. HGVS normalised.
Six specialised agents annotate, query databases, and mine literature in parallel.
Apply ACMG/AMP criteria systematically with per-rule confidence scoring.
Cancer risk, management recommendations, pharmacogenomic insights — with full provenance.
Every classification cites the underlying evidence with source, version, and retrieval time.
Six specialised agents orchestrated by LangGraph — every step, model call, and token logged.
Each agent owns a discrete job: VCF parsing, database annotation, literature retrieval, evidence ranking, ACMG criteria application, report generation.
Explicit state graph, not a chain of prompts. Branching, retries, and human approval gates are first-class.
Every classification is built from cards: PMID + DOI, ACMG criterion badge (PVS1, PS, PM, PP), study metadata, and a relevance score.
Eight first-class capabilities — no scripting required.
Drag-and-drop VCF, attach HPO phenotype terms, get pre-flight QC results.
Pathogenic, Likely Pathogenic, VUS, Likely Benign, Benign — with criteria trace.
Cite-anchored evidence extraction with PMID, DOI, cohort, effect size, p-value.
30+ curated gene-disease associations with inheritance patterns (AD, AR, XL).
Every claim shown as a card with source, version, ACMG badge, and relevance.
Real-time view of all 6 agents — status, token spend, retry budget, latency.
One query across variants, genes, diseases, publications — with filters and saved views.
Uploads, variants, evidence, agents, search — automate the workflow end-to-end.
Built around how genomic teams actually work — clinicians, researchers, counselors, and bioinformaticians each get their own view.
PHI never leaves your perimeter. Every model call is logged, replayable, and auditable.
Aligned to clinical-environment requirements out of the box. BAA available. Complete audit trail from upload to signed report.
On-premise or private-cloud deployment in your VPC / VNet / project. End-to-end encryption at rest and in transit. PHI never crosses the boundary.
Every agent step, LLM call, and classification decision is logged with timestamps, token counts, and model versions — replayable on demand.