Vidura Researcher AI: A Conversational Agent for Science Gateways
A usability study exploring how AI-powered search can help researchers and clinicians navigate science gateways more efficiently, overcome data overload, and access higher-quality information.
- Client: Innovation Diffusion Lab (IDL)
- Timeframe: 3 months (part-time)
- My Role: Quantitative Researcher, Data Analyst, Research Facilitator
- Team: Kerk Kee (PI), Ben Alfonsin, Ashraful Goni
- Methods: Usability Testing, A/B Testing, API Integration, Survey
- Tools: Qualtrics, Google Analytics, Zoom, SPSS, Excel, Figma
🪟 Project Overview
I joined the project in its final phase to guide the assessment of Vidura, an AI chatbot helping researchers find high-quality scientific evidence in health and neuroscience gateways.
- Measure user cognitive load when navigating complex research gateways.
-
Promote long-term, meaningful use of science gateways.
- Extend the diffusion and adoption of cyberinfrastructure tools through human-machine communication (HMC).
The platform was funded by two National Science Foundation grants with the goal of exploring how human-machine communication (HMC) can promote long-term, sustainable use of complex research infrastructure.
👩🏻🔬 Methodology
Design: Iterative usability testing in three gateway configurations
Participants: 15 Health professionals, medical students, and research staff
Tasks: Search for data fast, synthesize large datasets, assess information quality
Measurements: Task completion, time-on-task, clickstream analysis, survey data
Analysis: Nielsen User Satisfaction, NASA TLX, Modified SUS
🤿 Deep Dive
Prior testing (2022) used a competitive analysis between Google Scholar, the Gateway, and the chatbot. While informative, the design lacked focus on usability-specific pain points.
Proposal for Iteration
I proposed a new usability study targeting concrete problems: information overload, trust in chatbot guidance, and how user background shaped performance in real tasks.
I analyzed prior user journeys to identify common search behaviors like iterative refinement, citation validation, and quality vetting.
Realistic Task Framing
Tasks were aligned with user competencies (domain, technical, problem-solving) and grouped into three core challenges:
- Speed: “Find the most common COVID-19 vaccine in 2025.”
-
Scope: “Compile mRNA vaccine research into a synthesis.”
- Quality: “Compare high- vs low-quality COVID-19 treatment articles.”
Users were randomly assigned to use (1) the Gateway alone, (2) Vidura alone, or (3) both together.
Each condition allowed us to isolate or assess synergy between components.
Sample Size and Rationale
We used 5 participants per user group, following Nielsen’s benchmark for finding the majority of usability issues with small samples.
Google Analytics tracked page time, click patterns, and return behaviors. These were compared against industry standards (e.g., 14.6s average to find an item on Google).
Survey Instruments
We used:
- Nielsen User Satisfaction
-
NASA Task Load Index
-
NASA Modified System Usability Scale
💭 Reflection
- Conversational agents can enhance the functionality of complex platforms simplifying the management of massive amounts of data.
-
Effective usability testing depends on tasks being grounded in real user roles and cognitive patterns.
- AI assistants must be domain-aware and designed around researcher trust — especially when information quality is critical.
- I learned to adapt UX methods to expert-oriented systems where “user success” depends on deep domain knowledge.
- This project deepened my understanding of innovation diffusion and how AI tools fit into researcher workflows.
- Designing for medical researchers under cognitive strain taught me to value transparency, personalization, and search explainability in UX.