Ryght Case Study: Building a Life Sciences Generative AI Platform with Hugging Face
Ryght has officially launched Ryght Preview, an enterprise-grade generative AI platform designed specifically for the healthcare and life sciences sectors. By partnering with Hugging Face through the Expert Support Program, Ryght has developed a SaaS platform that provides industry-specific AI copilots and custom solutions to accelerate research, analysis, and documentation across complex data sources such as genomics, EMR, and clinical data.
Technical Architecture for Flexibility and Security
Ryght utilizes a "pluggable" LLM architecture to ensure the platform can evolve alongside the rapid development of medical-specific large language models. This design allows Ryght to seamlessly evaluate and integrate new specialized models as they emerge without disrupting existing services.
Integration of Hugging Face Inference Services
To achieve high performance and enterprise-grade security, Ryght integrated the following Hugging Face tools:
- Text Generation Inference (TGI) and Inference Endpoints: These services enable Ryght to register LLMs and link them to customer-specific inference endpoints. This setup secures connections and provides the flexibility to switch between different LLMs in real-time.
- Text Embeddings Inference (TEI): By serving open-source embedding models via TEI, Ryght reduced reliance on proprietary embeddings. This transition resulted in faster inference speeds, the removal of rate limit constraints, and the ability to deploy fine-tuned models tailored to the life sciences domain.
Performance and Scalability
To support multiple simultaneous customers and high volumes of concurrent requests, Ryght's infrastructure incorporates advanced batching, queuing, and distribution of model processing across GPUs. This architecture is designed to maintain low latency and prevent performance bottlenecks during high-demand periods.
Overcoming Development Challenges via Expert Support
Ryght leveraged Hugging Face's Expert Support Program to navigate the complexities of the AI landscape and accelerate their time-to-market.
Rapid Upskilling and Knowledge Transfer
Because the AI/ML field evolves rapidly, Ryght utilized asynchronous communication channels, regular advisory meetings, and technical workshops with Hugging Face experts to stay current on the latest techniques and models relevant to healthcare.
Strategic ML Selection
To avoid the "noise" of excessive tool and library options, Ryght worked with Hugging Face for solution design, proof-of-concept development, and production workload optimization. This partnership provided tailored recommendations on frameworks and models, streamlining the decision-making process.
Impact on Life Sciences Workflows
Ryght's platform aims to replace archaic data analysis methods that previously required large teams for simple queries or ML model development. The platform's copilot library is designed to accelerate:
- Information Retrieval: Swiftly extracting insights from diverse data sources.
- Synthesis and Structuring: Transforming complex unstructured data into actionable knowledge.
- Document Building: Reducing the time required for documentation from weeks to days or hours.
"Our partnership with Hugging Face's expert support has played a crucial role in expediting the development of our generative AI platform. The rapidly evolving landscape of AI has the potential to revolutionize our industry, and Hugging Face’s highly performant and enterprise-ready Text Generation Inference (TGI) and Text Embeddings Inference (TEI) services are game changers in their own right." — Johnny Crupi, CTO at Ryght