Deloitte needed to integrate large multimodal biomedical datasets, train custom protein and chemistry language models, and assemble drug-discovery pipelines far faster than manual infrastructure work allowed, because traditional drug development is slow, costly, and prone to failure.
Reported impact
Deloitte boosted developer productivity by 50 percent, saved 7–10 months of multi-node infrastructure work, reduced training runs from four weeks to eight hours, and shrank pipeline assembly from 4–6 weeks to a few clicks while enabling richer molecular ranking and generation workflows for drug discovery.
Deloitte built Quartz Atlas AI on NVIDIA DGX Cloud running on Oracle Cloud Infrastructure to turn multimodal drug-discovery research into a scalable, reusable pipeline. Atlas AI combines LLM-powered knowledge graphs, NVIDIA BioNeMo microservices, and Deloitte’s proprietary generative models to ingest sources such as PubMed, UniProt, antibody datasets, and small-molecule datasets into a 12-million-node, 97-million-edge graph totaling 5 TB of raw data. The platform uses AlphaFold2, OpenFold, and ESMFold for structure prediction, MegaMolBART and MolMIM for molecule generation, and DGX Cloud-based multi-node training to fine-tune protein and chemistry models such as a 15B-parameter ESM2 variant. NIM microservices then serve those models at scale so researchers can rank compounds, refine proteins, run inference loops, and generate comprehensive reports without manually wiring infrastructure or debugging bespoke pipelines for every experiment.
Scientific literature, protein, antibody, and small-molecule datasets are loaded into Quartz Atlas AI, DGX Cloud trains BioNeMo and Deloitte models on the multimodal graph, NIM serves structure and molecule models such as AlphaFold2 and MolMIM, and researchers receive ranked compounds, predicted properties, and generated reports for downstream wet-lab decisions.
By enabling us to easily run more concurrent experiments compared to our previous solution ... DGX Cloud has boosted developer productivity by 50 percent.
Deloitte is using NVIDIA BioNeMo models, available as NVIDIA NIM microservices, including AlphaFold2, OpenFold, and ESMFold for protein structure prediction, alongside MegaMolBART and MolMIM for molecule generation.
Our training runs have been reduced from four weeks to just eight hours.
We created a robust generative AI-powered knowledge graph with Atlas AI, loading over a dozen datasets, which amounts to 12 million nodes and 97 million connection edge links, totalling 5 terabytes in raw volume, searchable within seconds.
With DGX Cloud, multi-node training is now as easy as clicking a button, saving us seven to 10 months of infrastructure and tooling work.
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
NVIDIA Sample Feb 2, 2026 Leidos Leidos needed to reduce information overload and improve coordination in fast-moving disaster scenarios where human commanders must ingest video, radio, sensor, and public-safety signals quickly enough to allocate… NVIDIA AI Enterprise · NVIDIA Metropolis · NVIDIA NIM NVIDIA Sample Archived · Oct 5, 2024 ControlExpert ControlExpert needed to process more than 18 million motor claims a year across 30-plus countries, but manual review of vehicle photos, documents, invoices, and fraud indicators slowed down insurers and made consistent… more than 30 countries NVIDIA AI Enterprise · NVIDIA Triton Inference Server · NVIDIA-Certified Systems NVIDIA Sample Aug 2, 2023 BEN Group BEN Group needed a scalable way to analyze huge volumes of media, creator, and audience data so brands could place product integrations and influencer spend more effectively while also detecting fraud and improving… 100,000 Instagram posts analyzed per week NVIDIA DGX A100 