Accelerating Drug Discovery Research with an AI-Powered Chatbot on AWS
Industry: Biotechnology
Overview: A clinical-stage drug discovery and development company, specializing in predictive computational platforms for identifying novel drug targets and biological pathways for cancer immunotherapy treatments, faced challenges in efficiently retrieving insights from their extensive repository of internal research documents. These documents accumulated over years of experiments and needed an efficient solution to enhance knowledge management.
The company partnered with Commit, an AWS Partner, to design and build an AI-powered chatbot solution hosted on AWS. This chatbot allows researchers to ask natural language questions about internal experiments and retrieve relevant information, enabling them to gain insights more efficiently.
Solution Overview
The AI-powered chatbot provides researchers with precise and relevant insights from internal documents, focusing on two key requirements:
- Question Validation: Although initially considered, guardrails to validate the user’s input were deemed unnecessary due to the limited, internal nature of the user base.
- Context Enrichment using Retrieval-Augmented Generation (RAG): This technique enriches the user’s question with relevant contextual data from internal documents, ensuring responses are tailored to the researcher’s needs.
Commit designed the solution using AWS AI services as the core engine, relying on AWS Bedrock, which provides state-of-the-art large language models (LLMs) and RAG capabilities.
The architecture ingests documents from the company's Microsoft SharePoint environment whenever files are added, updated, or removed. API Gateway and AWS Lambda functions manage document synchronization, storing the files in an Amazon S3 bucket that serves as the knowledge base storage. The documents are periodically embedded and indexed using advanced vector similarity techniques with AWS Bedrock, making them searchable by the AI. The knowledge base is updated every 24 hours to ensure the most recent information is available.
When a user asks a question through the chatbot's web interface, the request triggers an AWS Lambda function that leverages the Bedrock agent to consult the knowledge base. It searches through internal documents and generates a natural language response, providing insights from the most relevant sources.
Comprehensive Conversation Tracking and User Feedback: The solution tracks all conversations, including chat logs, model outputs, user feedback, and performance scoring, stored in an Amazon DynamoDB database. Researchers can provide feedback on response quality, helping to refine the AI model’s performance. Administrators have full visibility into chat logs and metadata, enabling monitoring, model improvement, and report generation.
Continuous Improvement
Amazon Bedrock simplifies the adoption of advancements in generative AI, allowing quick transitions to improved models. This flexibility ensures the chatbot remains efficient and up-to-date with the latest AI capabilities.
Results and Benefits
Commit's solution, utilizing an advanced LLM model, improved response times and cost-efficiency while providing tailored recommendations. The company experienced several key benefits:
- Accelerated research velocity by enabling quick access to insights from years of internal knowledge
- Improved knowledge management and documentation practices
- Ability to extract more value from past research investments
- Centralized tracking of conversations for auditing and model refinement
- Scalable, serverless architecture to support growing usage
- Robust security controls leveraging AWS identity and access management
The success of this project has opened opportunities for the company to expand AI/ML capabilities on AWS, transforming how researchers interact with institutional knowledge and accelerating drug discovery and development.