Ben Alfonsin
About Me      


Evidence Pyramid Search Framework

A new way for medical professionals to gather life-saving research.

Industry: Medical
My Role: UX Operations, Research, and Design
Timeline: November 2023 - August 2025
Client: National Science Foundation
[End-to-End Design] [Usability Testing] [A/B Testing] [Survey] [AI]
Results: ~7 seconds faster than Google Scholar

Groundwork


Context Immunologists required quick access to medical research during the 2020 pandemic, but current tools made the research process time-consuming.Challenge
Current search engines require users to search based on tiles, dig through search results, and interpret jargon-heavy writing.
SolutionThe development of a new method for sorting research paper data facilitated through AI technology.

Process


[1] User ResearchPain points for medical research tools:

Speed: Quickly conducting research
Scope: Collecting many articles at once
Quality: Assessing methodological rigor


[2] Literature Review

Mental model match: Levels of Evidence PyramidIn medical research, papers are divided by types of evidence. This format is both familiar to researchers and addresses search engine shortcomings of speed, scope, and quality sortation.  

[3] High-Fidelity Prototype

10,000 immunology articles sorted by evidence level were used to create the high-fidelity prototype.
Through Google Dialogflow, an AI-enabled search was integrated into the page.

[4] User Testing
[4.1] Usability Testing
 20 users (medical professionals) were given 3 challenges.
Speed: Find an article within a time limit.
Scope: Write a summary based on a given research topic.
 Quality: Find articles with specific evidence levels.
[4.2] A/B Testing
Participants were given one of four tools to complete tasks.
Evidence Levels Sortation Only

AI Search Only
Evidence Levels with AI Search
Google Scholar

[4.3] Survey
 After tests, participants shared their feelings regarding:                               

1. Ease of Use
2. Satisfaction
3. Willingness to Adopt the Tool
[4.4] Results
  • Evidence Levels with AI Search performed the best. 7 seconds faster on average than Google Scholar.
  • Significant levels of usefulness, ease of use, ease of learning, satisfaction were reported by users of this version.
  • Participants discussed wanting better integration of search.

[5] Final Product
A new version which seamlessly integrates levels of evidence sortation and AI search within the same workflow.

Reflection


Future Iterations
  • More clear visualization of the selection of evidence levels.
  • Exploring the impacts of different AI models on output.


Learnings
  • AI technology cannot simply be applied to a product to improve it.
  • Rethinking standard work processes requires first assessing the user’s mental model.
  • The intentional integration of emerging technology requires an understanding of where its use may be beneficial.