deCypher: Dynamic Natural Language Querying for Engaging Insights

The BioPharma industry heavily relies on data to drive innovation, ensure regulatory compliance, enhance patient safety, inform business decisions, and improve healthcare outcomes. Utilizing data effectively has become a strategic imperative to optimize the results and efficiency of the Biopharma processes. However, a significant hindrance to this goal is the inadequacy of current data management systems in handling the complexity, diversity, and interconnectedness of data streams. Data is often siloed within convoluted datasets, placing the burden on average citizen data users to extract results and derive meaningful analyses. This places an unreasonable expectation on them to write complex queries reflecting an understanding of enterprise-wide data. Consequently, this strains their valuable time and hampers their ability to execute other meaningful tasks relevant to their roles successfully. Furthermore, existing systems fall short in assisting users with this problem, resulting in poor data utilization, lower-than-expected returns on data investments, and citizen data users spending an excessive amount of time on data administrative and processing tasks.

Semantic Querying

The DeCypher solution utilizes GPT-based models for accurate interpretation of the semantic meaning of natural language queries. The semantic meaning is then enriched with a contextual awareness that this is a product meant to handle the nuances of specialized medical terminology and complex pharma queries.

Query Optimization

The query is iteratively improved upon in consultation with the user. For the same, algorithms have been implemented for optimizing and refining queries, allowing users to receive precise and meaningful results efficiently. deCypher provides recommendations for query improvement based on user interactions and historical queries.

Data Integration

deCypher provides the capability to seamlessly integrate diverse data types, including genomic data, clinical records, and literature. deCypher accomplishes this by:

• Breaking down the query into data elements that would be required and making an informed estimate about the dataset where those elements would reside.

• These data elements reside in disparate data structures. deCypher supports multiple data structures commonly found in life sciences databases.

• These data elements differ in data types that would be returned from the datasets as well, some of them being images, texts, graphs; among others. deCypher has the ability to synthesize these data elements into composite insights.

Adaptive Learning

deCypher is by nature a learning and AI driven solution. Being a GPT model, it improves to adapt to evolving language use and user preferences over time. The learning is not limited to semantics but is incorporated in every layer of the solution, from query generation to data synthesis to better serve and adapt to the needs of the end user.

To summarize, deCypher provides an intuitive interface for interacting efficiently with complex datasets using natural language and extract actionable insights.

The solution primarily utilizes AWS Bedrock to access high performing foundational LLM models to interpret underlying datasets and generate contextualized responses. AWS OpenSearch Service / OpenSearch Serverless serves as the documents repository to store chunk sized information of Database Schemas, Data Dictionary etc. AWS Lambda is utilized to retrieve information from OpenSearch and connect with Bedrock to generate responses.