Cloud for Scientific AI 
and Healthcare

CRISPR-GPT is an LLM-powered agent system developed by scientists from Stanford, Princeton and Google DeepMind to automate gene editing experiments, from CRISPR system selection to sgRNA design and data analysis.

Goal: Transform gene editing from a months-long process into automated workflows accessible to any scientist.

Solution: Enabling rapid model screening and fine-tuning via Nebius.

Result: Junior researchers with no gene editing experience now achieve 80-90% efficiency on first attempt. Undergraduate students are onboarded in a day, and experts work faster by using AI agents to help run analysis, check designs and troubleshoot experiments.

Unum scales IO efficiency in AI-driven life sciences workloads with StringZilla — an open-source, high-speed string processing library that accelerates computationally heavy tasks like hashing, sorting, and fuzzy-matching.

Goal: To fully leverage HPC architectures by optimizing the software layer.

Solution: Enabled by Nebius, Unum ported StringZilla to GPUs using hardware-specific kernels to tap into parallel processing capabilities for more efficient string scoring.

Result: Delivered orders-of-magnitude speedups for critical sequence analysis tasks like DNA, RNA, and protein alignment and fingerprinting.

Slingshot AI is building the first foundation model for psychology, reimagining mental health AI with its app, Ash. Trained on the largest behavioral dataset, Ash personalizes each journey via fine-tuning and RL.

Goal: Develop a responsible AI for scalable and tailored mental health support.

Solution: Based on real health data, Ash’s 32B-235B models were trained on fast Nebius GPU clusters with DeepSpeed and Zero-3. Fine-tuning and RL used expert-guided reward models.

Results: Adapting to users through reinforcement learning, the app opens a new in-silico therapy modality. A recent NYU study found that Ash helps users feel more connected, countering the idea that AI isolates people.

Helical builds Virtual AI Labs for pharma and biotech teams to scale virtual experiments in hours, not months. Their flagship model, Helix-mRNA, captures long-range biological signals and can be quickly adapted to proprietary data.

Goal: Develop personalized, production-grade Bio FMs for drug discovery pipelines.

Solution: Helical relies on Nebius to pretrain and post-train Helix-mRNA — a 500M-parameter hybrid model combining Mamba-2, transformers and codon-aware tokenization.

Results: Model training completed in 10 days, followed by 36-hour specialization runs per dataset. Helix-mRNA outperformed prior baselines, improving correlation scores by up to 146%. With Nebius, Helical scales faster and delivers in-silico tools in weeks — not months.

xAID develops a foundation model for chest and abdomen CT scans, addressing the need for accurate diagnostic support amid a global radiologist shortage.

Goal: Train a large 3D transformer model on volumetric CT data, to detect 70+ pathologies across body regions.

Solution: Run compute-heavy training on large 256³ voxel CT scans by using FP16 precision and accumulated gradients.

Result: Strong F1 scores, stable multi-day training cycles and scalable model development.

SieveStack is building the world’s largest dataset of molecular simulations to train a multi-layered stack of foundational models and advance dynamics-driven drug discovery.

Goal: To unlock treatment pathways for hard-to-treat diseases with AI-powered, physics-based modeling and biochemistry.

Solution: To generate high-precision molecular dynamics simulations and optimize model training with a mixed-precision approach — FP32 for accuracy and BF16 for performance.

Result: SieveStack leveraged Nebius and TractoAI support team to prototype, debug, and scale foundational models with >90% GPU utilization, revealing drug-target interactions beyond the reach of lab experiments.

Converge Bio is a biotech company pioneering the use of LLMs to analyze single-cell RNA sequencing data. Their work aims to transform how scientists understand disease mechanisms and therapeutic responses at the individual patient level.

Goal: To develop a scalable, high-performance foundation model capable of analyzing raw gene expression data across 20,000+ genes per cell and delivering explainable, patient-specific insights for drug discovery and precision medicine.

Solution: Converge trained Converge-SC on over 36 million cells and 2TB of data, leveraging H100 80GB GPUs and advanced parallelism techniques. The model retained full numerical fidelity, operated at the patient level, and was built for interpretability and usability.

Result: Converge-SC outperformed baseline models across disease classification and drug perturbation tasks. Now released via Hugging Face, the model is helping biotech and pharma teams unlock insights faster — bringing explainable AI to the forefront of biomedical research.

YerevaNN strives to fast-track generative molecular drug design. By leveraging molecular language models, the research center enables the precise, code-driven engineering of pharmaceutical compounds.

Goal: Optimize drug candidates with three continuously pre-trained models integrated by a genetic algorithm.

Solution: YerevaNN streamlined implementation with Flash Attention, FSDP, and post-backward prefetch for enhanced parallelism and faster preprocessing and training.

Result: With Nebius, YerevaNN fast-tracked text tokenization, expedited docking runs for molecular modeling, and improved performance by 8% over previous methods, processing 180,000 words per second.

Simulacra AI is transforming the quantum chemistry field by automatically generating high-precision datasets for molecular dynamics models at scale.

Goal: Build a scalable foundational wave-function model for molecular systems that can generate high-accuracy datasets for pipelines of drug and material discovery.

Solution: Simulacra AI used Nebius infrastructure to overcome scalability and efficiency challenges.

Result: Simulacra AI delivers next-generation molecular data, enabling any company to refine in silico pipelines without relying on broad internal infrastructure to train models.

Quantori is the end-to-end data, technology and digital services partner of choice for leading biopharma and healthcare organizations worldwide.

Goal: To develop an AI framework that generates molecules with precise 3D shapes, enhancing drug discovery and material design.

Solution: Quantori employs a pipeline based on Equivariant Diffusion Model and Structure Seer model trained on 1.6M molecules from the ChEMBL database. The pipeline generates molecular structures using shape descriptors.

Result: After 1,500 training epochs, the model successfully generated chemically sound molecules that closely resemble real molecules in shape. The approach enables rapid molecular ideation, predicting valid 3D conformations with optimized properties.

TrialHub is a data intelligence platform designed to make clinical trials more efficient and patient-centric.

Goal: To deliver quantifiable insights from unstructured, text-heavy data — scaling to production in days.

Solution: To launch an MLOps-optimized vector embedding pipeline with NVIDIA L40s and Nebius expert support.

Result: Deployed one of the largest vector databases in clinical research, with 250 million vectors, reducing trial delays and amendments by half.

Lynx Analytics delivers AI-powered solutions to enterprises across life sciences, telecom, retail and finance. Its platform, LynxKite 2000:MM, lets experts build predictive Graph AI workloads to enhance reasoning without writing code.

Goal: Power fast and flexible Graph AI workflows.

Solution: Using Nebius’ VMs and managed Kubernetes to scale workloads, LynxKite 2000:MM automates container orchestration, maintaining high utilization and rapid deployment.

Results: With Nebius, Lynx Analytics accelerates delivery across diverse AI projects. Easy provisioning and Slurm support via Soperator enable engineers to focus on modeling, while flexible scaling keeps performance and costs optimized.

Compugen is a clinical-stage biotech company pioneering AI/ML-powered drug discovery for cancer immunotherapy. Its Unigen™ platform integrates multi-omics, spatial, and single-cell data to identify new drug targets and resistance pathways.

Goal: Develop AI models that uncover new immune features in cancer datasets.

Solution: Nebius helps train large models, including single‑cell RNA transformers. Models up to 3B parameters run on BF16/FP32 with minimal setup, thanks to Nebius’ ease of use.

Results: The team trained a unique model that predicts spatial features (e.g., TLS) from non-spatial data, enabling deeper analysis across broader datasets. Nebius provided flexibility, performance, and simplicity, helping Compugen integrate AI into its Unigen discovery pipeline.

Sword Health used Nebius to help bring Dawn to life, expanding its AI Care model into direct-to-consumer mental wellbeing support. The company combined years of clinical AI development with new inference architecture to make long, sensitive conversations safer, faster and more scalable.

Goal: Build a mental health wellbeing system that could deliver clinically rigorous, high-quality support over long, multi-session conversations while meeting strict safety, governance, latency and cost requirements.

Solution: Sword paired MindGuard and MindEval with Nebius Token Factory and dedicated NVIDIA Blackwell endpoints. Using custom speculative decoding, KV caching and prefix-aware routing, the team moved from a 30B model to a 200B+ parameter model without sacrificing production responsiveness.

Result: Sword reduced tail latency from more than 20 seconds to under 12 seconds, cleared its quality bar for Dawn, and created a production-ready platform for AI wellbeing support at scale. Nebius also powered Sword’s broader AI footprint, including Thrive and high-volume AI-guided care delivery.