Modern hospitals and public health systems need continuous energy, logistics, and procurement visibility. Within the DWS IQ Industrial Intelligence suite — core engine DWS IQ 6, dual-use tier Aegis — the CareGreen Agent turns clinical operations into audit-ready evidence for the healthcare regulatory stack: CSRD (ESRS E1 + ESRS S1), GDPR, EHDS, and MDR.
Healthcare estates, labs, and supplier networks are covered across energy, emissions, social, and governance requirements. DWS IQ connects facility, procurement, and clinical data so reporting is always current — the ontology DWS IQ uses to serve the healthcare vertical (facility carbon, ESRS S1, EHDS-aligned secondary data).
2025 reference: the EU Health Data Space (EHDS) Regulation entered into force in 2025, with secondary-use provisions phasing in over the following years; the first wave of CSRD sustainability statements for FY2024 was filed in 2025 under ESRS.
Automate carbon tracking for HVAC, district heating, and emergency power with real-time dashboards.
Track pharma, equipment, and consumable emissions across vendors to satisfy CSRD supply chain scopes.
Plan fleet transition timelines, costs, and emissions reduction milestones in one shared view.
Prioritize retrofits with ROI + CO₂ impact models to meet Fit for 55 timetables.
⚠ Illustrative / representative · pre-GA. Anonymized walkthrough of how the ontology behaves. Not a real hospital, patient, incident, or consented deployment. No real patient or clinical data is implied.
A representative regional hospital group (anonymized) connects facility and procurement data to DWS IQ 6. The trace below shows trigger → agents → regulatory outcome.
June 2026 · BIO 2026. NVIDIA announced BioNeMo Agent Toolkit — domain-specific tools and skills for agentic life sciences. DWS IQ integrates BioNeMo skills to bring accelerated scientific computing to healthcare workflows.
The NVIDIA BioNeMo Agent Toolkit gives AI agents the tools needed to synthesize scientific knowledge, call models, evaluate results, and execute next actions across biology, chemistry, genomics, and drug discovery. More than 50 leading companies — including Anthropic, OpenAI, Lilly, Schrödinger, Databricks, and Snowflake — are adopting the toolkit.
Agents identify small-molecule drug candidates by generating compounds, docking them to targets, predicting binding strength, and filtering for drug-like properties — compressing screening timelines from days to minutes.
NVIDIA Parabricks accelerates alignment and variant calling; genomic foundation models score variant effects and rank the most disease-relevant candidates for further study.
Agents design and validate protein binder candidates computationally before lab work begins, compressing traditionally labor-intensive design cycles.
Agents connect real-world data to reasoning models for literature review, protocol generation, clinical trial screening, and pharmacovigilance.
Powered by NVIDIA NIM microservices, Nemotron reasoning models, NemoClaw blueprints for secure private agents, and the OpenShell runtime for controlled execution — all integrated into the DWS IQ sovereign EU-hosted platform via agent-callable skills.
DWS IQ integrates the best tools from NVIDIA, Google DeepMind, and the open-source ecosystem — curated per industry vertical. Each toolkit is agent-callable via NIM microservices or containerized pipelines on sovereign DGX Spark infrastructure.
End-to-end AI platform for healthcare: imaging, genomics, and medical devices. The umbrella that connects BioNeMo, MONAI, Holoscan, and Parabricks into a unified development environment.
Open-source medical imaging AI framework — 50+ pretrained models for CT, MR, pathology, and endoscopy. Supports 2D/3D segmentation, registration, reporting, and multimodal workflows. PyTorch-based.
Real-time runtime for GPU-accelerated medical devices. Ingests high-bandwidth, multimodal sensor data, performs AI reasoning, and drives deterministic actions — surgical robotics, ultrasound, endoscopy.
Federated Learning Application Runtime Environment. Trains AI models across hospital networks without sharing raw patient data — GDPR and EHDS compliant. Used by Mayo Clinic, Mass General, and NCI.
GPU-accelerated genomics pipeline. Alignment, variant calling, and annotation that runs 30–50× faster than CPU-only tools. GATK-compatible output for clinical and research workflows.
Agent-callable microservices for protein structure prediction, molecular docking, generative chemistry, and biomarker discovery. The agentic life sciences toolkit announced at BIO 2026.
State-of-the-art protein structure prediction — proteins, nucleic acids, ligands, and their interactions. Foundation for understanding disease mechanisms and drug–target binding across oncology, autoimmune, and infectious disease.
AI-driven protein binder design. Generates novel high-strength binders for cancer targets (VEGF-A, PD-L1, TrkA), viral proteins (SARS-CoV-2), and autoimmune disease markers with surgical precision.
DeepMind drug discovery spin-off. Applying AlphaFold-class models to pharmaceutical drug design — advancing beyond structure prediction into end-to-end molecular design and optimization.
Open-source, trainable reimplementation of AlphaFold. Trained on 300K+ experimental structures and 13M+ synthetic structures. Allows custom fine-tuning on proprietary datasets — critical for sovereign deployment.
Industry-standard workflow orchestration for bioinformatics. 83+ curated pipelines (nf-core) for genomics, transcriptomics, and proteomics. 43% citation share in bioinformatics WMS. Reproducible, scalable, publishable.
Gold-standard variant calling. Broad Institute GATK for germline/somatic variants; Google DeepVariant (available as NVIDIA NIM) for deep-learning-based variant calling with higher accuracy on complex regions.
Foundational genomics utilities for reading, writing, editing, indexing, and viewing SAM/BAM/CRAM files. Core dependency of every modern sequencing pipeline from alignment to variant annotation.
Platform pattern: Every DWS IQ industry vertical ships a curated toolkit (models + libraries + pipelines) paired with sovereign infrastructure (DGX Spark + DWS Rooms). The toolkit runs as agent-callable skills; the infrastructure keeps data on-premise and audit-ready.
Cloud-only LLMs break on jurisdiction, logging, and control. Healthcare data — patient records, genomic sequences, clinical trial outputs — demands sovereign, on-premise AI infrastructure with full audit trails and zero U.S. cloud dependencies in the critical data path.
GB10 Grace Blackwell desktop AI workstation — 128 GB coherent unified memory, 1 PFLOPS FP4, <1 kW power. Runs all DWS IQ agents, BioNeMo NIM microservices, NemoClaw sandboxed runtime, and Ollama models on-site with no data leaving the hospital LAN.
Strategic command center for hospital leadership — 2× Acer Veriton GN100 (bridged) + NVIDIA model bundle. Portfolio-level visibility across revenue, EBITDA, ESG rating, Fit for 55 compliance, and CSRD readiness for board members and CxO executives.
Operational command center for day-to-day clinical and facility management — multi-site dashboards, real-time status, workforce management, equipment tracking, safety management, and environmental compliance coordination across hospital networks.
Perception → Imagination → Action. Real-world sensor ingestion (IoT, BMS, medical devices), NVIDIA Cosmos world-model simulation for forecasting outcomes, and agentic loop engineering for autonomous remediation — all behind human-in-the-loop gates.
Schedule a review and get a sector-specific roadmap for Fit for 55 and CSRD readiness.
Book a Healthcare Review
Every DWS IQ agent ships with executable skills — invoked via
/skill-name
slash commands or the MCP
tools/list
API. Skills are the product: reusable workflows, hard compliance gates, and
agent-accessible capabilities your team can run today.
Hospital facility carbon + pharmaceutical Scope 3. Tracks medical gases, sterilization energy, procurement.
Deep research across EUR-Lex, EFRAG, and national MDR/IVDR guidance. Cited sources, confidence scores.
Runtime watchdog. Streams telemetry, detects anomalous agent behaviour in patient-data-adjacent workflows.
European Sovereignty Compliance Climate Intelligence Platform
Bring your multi-agent teams and enterprise data to a sovereign EU-hosted platform. Each
industry gets its own model evaluation — the
AgentConfig.model
field is per-agent, running a trained Lifetime World Model + open LLM with continuous
benchmarking. The Lifetime World Model is the enterprise learning layer that fine-tunes
your data and regulations, optimizes market value, and ensures competitive advantage
through multi-agent intelligence tools.
Gemma 2 27B (EU-hosted)
EU-hosted deployment mandatory for patient-adjacent data. Compact model suitable for edge inference in hospital environments. Strong at medical waste classification and pharmaceutical supply chain.
Connect your existing AI agents to DWS IQ via the A2A protocol. Your agents publish an Agent Card, and our platform discovers and orchestrates them alongside the CareGreen Agent.
Every BYO agent is bound to a verified human identity, isolated in a MicroVM sandbox, and subject to TC-4 Leash-Snap termination. We aim to open-source the KYA Standard via the Agentic AI Foundation, NIST, and European standards bodies. Know Your Agent Standard →
Example: HUS brings facility management agents + pharmaceutical procurement agents for Scope 3 supply chain tracking
The Lifetime World Model is the enterprise learning layer that adds intelligence on top of your chosen LLM. It fine-tunes enterprise data and regulation, optimizes market value, and ensures competitive advantage with multi-agent tools — trained specifically on your industry context.
Dual-use angle: Field hospitals, CBRN medical, defence medical supply
Military and civil protection healthcare — field hospitals, CBRN decontamination, medical supply chains for defence — demands sovereign AI. Aegis provides Article 12-logged medical AI decisions, export control for dual-use medical equipment, and ICEYE SAR monitoring of healthcare infrastructure in contested or disaster areas.
CSRD, EU AI Act (high-risk), MDR, Dual-Use Reg. 2021/821
Healthcare — page v1.4 · 2026-06-27 · Toolkit & Libraries (NVIDIA + DeepMind + Open Source)