Data Platform AI
The consented, de-identified data layer that powers the Mergic flywheel. 5+ modalities unified, petabyte-scale, sub-second retrieval.
Product capabilities
Atlas powers all Mergic products with a consented, de-identified data layer that merges imaging, genomics, pathology, EHR, and lab data — petabyte-scale with sub-second retrieval for research and model training.
Build unified patient graphs that link imaging studies, genomic variants, pathology slides, clinical notes, and lab values across time. Query cohorts by cross-modal phenotypes — "stage III NSCLC with EGFR exon 19 deletion and CT-detected pleural effusion" — and assemble training datasets in minutes instead of months.
Automated PHI scrubbing, consent tracking, and data-use agreements — compliant with HIPAA and IRB protocols.
GPU-accelerated vector embeddings for imaging, genomics, pathology, and text — unified semantic search across modalities.
Full provenance for every data point — source system, ingestion time, transformations, and downstream model usage.
Train models across institutional boundaries without moving PHI — federated learning with privacy guarantees.
GPU-accelerated similarity search across imaging, genomics, pathology, and text — find similar patients by phenotype.
NVIDIA healthcare stack
Petabyte-scale multimodal data storage with GPU-accelerated indexing, vector search, and cross-modal retrieval.
Every clinical decision and wet-lab result flows back into Atlas for continuous model improvement with DGX training.
Performance metrics
Modalities unified — imaging, genomics, pathology, EHR, lab values merged into longitudinal patient graphs
Storage and retrieval infrastructure for multimodal health data — imaging, genomics, pathology at scale
Sub-second cross-modal queries across millions of patients — GPU-accelerated vector search and retrieval
Patient graphs spanning years of imaging, genomics, pathology, and clinical history — unified temporal views
Data lineage tracking — full provenance from source systems to model training and clinical decisions
Use cases
Assemble training datasets by querying cross-modal phenotypes — "metastatic breast cancer with HER2 amplification and liver metastases on CT" — across millions of de-identified patients. Atlas returns longitudinal patient graphs with imaging, genomics, pathology, and EHR in seconds. Researchers export cohorts with full data lineage for model training, clinical trial recruitment, or retrospective studies. Every data point is consented, de-identified, and auditable.
Curate high-quality training datasets for Vision, Helix, Lumen, and Forge — de-identified, consented, lineage-tracked.
Track disease progression, treatment response, and imaging changes over months to years — unified temporal graphs.
Find patients similar to an index case by cross-modal phenotype — imaging features, genomic alterations, pathology findings, and clinical presentation. GPU-accelerated vector search across all modalities returns similar patients in sub-second query time.
Every clinical decision and wet-lab result flows back into Atlas for model retraining — continuous improvement loop.
How it works
Atlas unifies imaging, genomics, pathology, EHR, and lab data into consented, de-identified longitudinal graphs — powering research, model training, and clinical decision support.
Pull imaging (PACS/VNA), genomics (FASTQ/VCF), pathology (whole slide), EHR (FHIR), labs from source systems
Automated PHI scrubbing, consent tracking, and data-use agreement enforcement — HIPAA and IRB compliant
Build longitudinal patient graphs linking all modalities across time — unified by patient ID and temporal events
GPU-accelerated cross-modal embeddings for imaging, genomics, pathology, and text — unified semantic space
Petabyte-scale storage with full data lineage tracking — source system, ingestion time, transformations
Sub-second cohort queries, similarity search, and longitudinal views — powers all Mergic products
Integration
Ingest DICOM imaging and whole-slide pathology from PACS, VNA, and digital pathology platforms
Ingest FASTQ, BAM, VCF from sequencing platforms and variant databases — ClinVar, gnomAD, OncoKB
Connect to Epic, Cerner, and other EHR systems via HL7 FHIR or custom connectors for clinical data ingestion
Related products
Real-time GPU radiology and pathology AI for detection and segmentation.
Whole genomes from a day to under an hour with Parabricks-backed secondary analysis.
The clinical-decision agent that sees across modalities — imaging, genomics, EHR.
Generative AI for protein structure prediction and molecular design with BioNeMo.
HIPAA-compliant deployment with full audit, model versioning, and IP protection.
FAQ
Answers for research teams, data scientists, and IT leaders evaluating Atlas for multimodal health data.
Atlas automates PHI scrubbing using industry-standard de-identification algorithms, tracks patient consent status and data-use agreements, and enforces IRB protocol compliance. All data is auditable back to source systems with full lineage tracking.
Imaging (DICOM radiology, whole-slide pathology), genomics (FASTQ, VCF, variant annotations), EHR (FHIR clinical notes, diagnoses, medications), lab values, and any structured or unstructured health data.
Yes. Atlas supports federated learning workflows where models train across institutional boundaries without moving PHI. Each site runs local training on its Atlas instance, and only model gradients are shared.
Every clinical decision (from Lumen), imaging annotation (from Vision), variant call (from Helix), and wet-lab result (from Forge) flows back into Atlas with full provenance. This continuously growing dataset powers model retraining with DGX infrastructure for perpetual accuracy improvement.
Get started
Bring a research cohort, model training, or longitudinal patient view use case. We'll map the data unification, cross-modal queries, and first measurable outcome.