LynxKite 2000:MM | Graph AI Platform

Skip to content

Ask LynxScribe

LynxScribe

Start new conversation

LynxKite is a powerful tool built by Lynx Analytics to create graphs from very large datasets and to use graphs as functional components of larger AI workflows. The latest version, LynxKite 2000:MM, has been developed to accelerate graph computation with cuGraph to support Gen AI and Graph RAG applications. It has been optimised to run on NVIDIA GPUs. Unlike graph databases such as neo4j and graph computation frameworks like Spark, data stays on the GPU for maximum performance.

When combined with NVIDIA’s BioNeMo SDK, RDKit, and Graph Neural Networks (GNNs), LynxKite 2000:MM presents a powerful solution for the pharmaceutical industry. By integrating these workflows, pharmaceutical companies can uncover novel insights into drug discovery, target identification, and molecular interactions.

Native Support for NVIDIA GPU Clusters: Optimized to run on NVIDIA GPUs with a fallback CPU mode for flexibility.
Integration with cuGraph Libraries: Harnesses industry-standard GPU-accelerated graph analytics.
Integrating NVIDIA BioNeMo: Leverages pre-trained generative AI models that use molecular representations (e.g. SMILES).
Integrating RDKit: RDKit can serve as a tool to validate and refine molecules generated by BioNeMo's generative models.

Over 600 Graph Algorithms, Including NetworkX: Doubling the capabilities of previous LynxKite versions. More than 100 algorithms are accelerated by cuGraph.
Multi-User Collaboration: Enabling teams to work together efficiently on complex projects.
Task-Specific Workspaces, such as agentic LLM logic flow development (including NVIDIA NIMs), chatbot development (powered by LynxScribe) and Graph Neural Network (GNN) architecture design.
Seamless Data Format Conversion: Automatic adaptation between formats for effortless Python tool integration.

With LynxKite 2000:MM you can construct graph-based retrieval workflows that enhance LLMs with structured, context-aware reasoning. Instead of relying on flat text retrieval, LynxKite 2000:MM stitches together Graph RAG pipelines, where document fragments are organized into a knowledge graph, enabling smarter, relationship-driven retrieval. By leveraging graph traversal and link prediction, LynxKite 2000:MM dynamically retrieves the most relevant and interconnected information for a given query. This structured retrieval ensures that LLMs receive precise, semantically rich context before generating a response. The result? More accurate, nuanced, and context-aware outputs, making it ideal for complex query resolution, personalized recommendations, and predictive analytics in enterprise AI applications.

With LynxKite 2000:MM, users can design, assemble, and run advanced RAG systems enhanced by Graph AI. User queries and workspace interactions generate structured Y.js documents, which are dynamically analyzed through graph-based retrieval. Unlike traditional RAG, which relies solely on embedding proximity, LynxKite 2000:MM enables the integration of Graph Analytics (via cuGraph, NetworkX) to uncover deeper contextual links. By seamlessly combining Gen AI with graph traversal, the system retrieves semantically relevant but structurally distant nodes, delivering more accurate and context-aware responses in chat applications.

LynxKite 2000:MM simplifies complex medical research tasks, such as classifying cell types or predicting disease states, by providing an intuitive workflow environment for analytical processes. Medical teams working on preclinical or clinical workflows can seamlessly integrate multiple data sources, libraries, and models to create unified AI workflows. The platform enables users to visualize, manage, and even combine workflows effortlessly. By offering a user-friendly management layer, LynxKite 2000:MM makes advanced frameworks like BioNeMo and Graph Neural Networks more accessible to a broader audience. Additionally, it empowers superusers to streamline the management and assembly of data, models, and computational components, enhancing collaboration and accelerating research efficiency.