Workflow · Production

Aetion uses Amazon Bedrock to translate natural language scientific intent into patient variable measures

The problem

Scientists, epidemiologists, and biostatisticians had to manually translate complex natural language scientific intent into algorithmic patient variable definitions (Measures), a process requiring specialized training and days of effort.

Workflow diagram · grounded in source
1
User submits natural language query
trigger
“When a user asks a question through the assistant UI, Substantiate initiates a request containing the question and previous history of messages, if available”
2
Semantic RAG retrieval
ai_action
“Questions and answers are dynamically selected from a local knowledge base based on semantic proximity to the user question. These examples improve the quality of the generated answers by incorporating similar previously asked and answer…”
3
LLM generates measure instructions
ai_action
“Measures Assistant incorporates the question into a prompt template and calls the Amazon Bedrock API to invoke Anthropic's Claude 3 Haiku”
4
Guardrails validation
validation
“Measures Assistant maintains a local knowledge base about AEP Measures from scientific experts at Aetion and incorporates this information into its responses as guardrails. These guardrails make sure the service returns valid instruction…”
5
Instructions delivered to user
output
“A user asks Measures Assistant a question expressed in natural language and receives instructions on how to implement this”
6
Expert Q&A pool refinement
feedback_loop
“The generation and maintenance of the question-and-answer pool involve a human in the loop. Subject matter experts continuously test Measures Assistant, and question-and-answer pairs are used to refine it continually to optimize the user…”
Reported outcome

Measures Assistant enables users to convert natural language scientific questions into AEP Measures in minutes rather than days, without requiring support staff or specialized training.

Reported metrics
Time to implement measuresminutes as opposed to days
Specialized support requiredwithout the need of support staff and specialized training
Reported stack
Amazon BedrockClaude 3 HaikuKubernetes on AWSmxbai-embed-large-v1
Source
https://aws.amazon.com/blogs/machine-learning/how-aetion-is-using-generative-ai-and-amazon-bedrock-to-translate-scientific-intent-to-results?tag=soumet-20
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Measures Assistant enables users to convert natural language scientific questions into AEP Measures in minutes rather than days, without requiring support staff or specialized training.

What tools did this team use?

Amazon Bedrock, Claude 3 Haiku, Kubernetes on AWS, mxbai-embed-large-v1.

What results were reported?

Time to implement measures: minutes as opposed to days; Specialized support required: without the need of support staff and specialized training (source-reported, not independently verified).

How is this workflow AI workflow structured?

User submits natural language query → Semantic RAG retrieval → LLM generates measure instructions → Guardrails validation → Instructions delivered to user → Expert Q&A pool refinement.