Technology
How Omnis Genomics is built
A GPU-native stack with verification baked in. We share the details because serious teams need to know how the plumbing works before they trust it.
GPU first from day zero
- • Custom kernels for FM index mapping, Pair-HMM, and streaming ingest.
- • Continuous streaming instead of CPU↔GPU batch ping-pong.
- • Multi-GPU scheduling keeps devices saturated while managing artifacts.
Reproducibility by design
- • Golden datasets and benchmark harnesses for every release.
- • Deterministic configurations with pinned containers and SBOMs.
- • Artifact lineage preserved across pipelines and exports.
Verification layer: VeriBiota
- • Truth-set comparisons, drift detection, and delta analysis per run.
- • Machine-readable reports for audits and collaborators.
- • Hooks for CI so pipelines ship only after verification gates pass.
Example: Nextflow wrapper into OGN
What it looks like to swap a CPU step with the GPU rail. Deterministic container, verification gate, and artifact export defined in one block.
process OGN_WGS {
container 'ghcr.io/omnis/ogn-wgs:stable'
input:
path reads1
path reads2
script:
"""
ogn submit --reads ${reads1},${reads2} --pipeline wgs-30x --export vcf,metrics.json --verify veribiota:giab-hg002
"""
}Roadmap
Q1
- • Multi-sample calling
- • Expanded structural variant modules
Q2
- • Deeper Helix Studio ↔ OGN integration
- • Artifact embeddings for ML/LLM stacks
Q3
- • Additional cloud + on-prem blueprints
- • Status page with real-time fleet telemetry