Bernie - AI Deep Research Assistant for Biopharma Intelligence
Designing BioCentury’s first AI-native research experience to turn complex, fragmented biopharma intelligence into structured, source-backed research.
Goal
Enable biopharma professionals to turn complex research questions into credible, actionable intelligence efficiently, reducing the effort needed to synthesize fragmented information while maintaining transparency and control for critical decisions. For BioCentury, Bernie establishes a foundation to expand AI-powered capabilities across the platform and create new opportunities for differentiated, higher-value research.
Role
Principal UX Designer
Focus
UX Strategy · Information Architecture · AI Experience Design · Interaction Design
Overview
Bernie was a BioCentury initiative designed to support strategic research across scientific, clinical, competitive, and investment domains. Its foundation combined BioCentury’s structured intelligence, editorial content, and domain expertise with broader research sources.
The May 2025 strategy brief identified the initiative’s stage as Proof of Concept and Infrastructure Development. It proposed an internally tested Alpha focused on three use cases: Horizon Scanning, Competitive Landscape Analysis, and Clinical Trial Strategy.
I translated the product direction into an experience architecture and wireframes, organizing the process into four stages: Refine Research, Select Sources, Extract Data, and Generate Report. Each stage served as a checkpoint for researchers to guide the inquiry or review supporting material.
The Challenge
Turning fragmented intelligence into actionable answers
Biopharma professionals analyze clinical trials, pipelines, scientific literature, regulatory activity, company intelligence, and deal data to answer questions that influence development, investment, and partnership decisions. For example, “What are the most promising unpartnered assets in inflammation?” requires interpreting scientific findings, assessing clinical progress, and linking commercial signals to strategic objectives.
Even with premium intelligence platforms, this work requires extensive searching, cross-referencing, and synthesis. BioCentury refers to this friction as “decision drag”: the gap between needing an insight and having enough context to act confidently.
The opportunity was to use AI to reduce this effort. The UX challenge was to accelerate research while keeping the question, source choices, and supporting evidence visible, ensuring researchers maintained meaningful control over the inquiry and its supporting analysis.
The research task spans multiple intelligence domains. Connecting those sources through research, cross-referencing, and synthesis frames the work Bernie was designed to support.
The Approach
Structuring research around visible checkpoints and researcher judgment
Complex research questions are often underspecified, draw on multiple sources, and require decisions about which evidence should inform the analysis. My approach addressed these needs through a guided interaction model.
I structured Bernie around four progressive stages:
Each stage assigned a specific research activity to the interface: defining the question, choosing sources, inspecting findings, and building a report. This allowed researchers to understand the system’s actions and guide the inquiry before it progressed.
The wireframes implemented this architecture from research initiation through report output. They also addressed stage transitions by displaying work in progress, summarizing completed steps, and highlighting upcoming activities.
Refine Research → Select Sources → Extract Data → Generate Report
CAPS TITLE
A RESEARCH WORKFLOW WITH VISIBLE CHECKPOINTS
Four stages make researcher participation part of the proposed AI workflow.
The architecture established checkpoints where choices could shape the research, such as clarifying scope, selecting inputs, and reviewing evidence before synthesis. Progress indicators and completed-stage summaries helped maintain context as the inquiry developed.
01. Refining the Research Question
Address ambiguity before generating answers
A question like “What are the most promising unpartnered assets in inflammation?” contains assumptions that can significantly affect the research. For example, “unpartnered” may refer to various ownership structures or exclusivity arrangements, while “promising” leaves evaluation criteria undefined.
I designed the refinement experience to highlight this ambiguity. The interface explained the importance of clarification and presented suggested formulations alongside the original question.
Researchers could select a suggestion, edit the inquiry, or continue with the original wording. This supported defining a more focused question while preserving the researcher’s authority over its meaning and scope.
The original question remains visible alongside suggested refinements and an explanation of why greater specificity could help.
02. Giving Researchers Control Over Sources
Make the evidence behind the research explicit
Strategic biopharma questions require information from scientific, clinical, regulatory, commercial, and private research sources. I made source selection a distinct checkpoint, letting researchers review proposed inputs and decide which should contribute to the analysis.
The wireframe grouped sources into specialized sources, general sources, and a researcher’s private library. Within the specialized group, BioCentury’s library appeared alongside sources such as ClinicalTrials.gov, conference abstracts, and regulatory agencies. Selection controls and counts translated this broad information landscape into concrete research choices.
This structure allowed researchers to shape the evidence base before extraction. The source list represented the intended interaction model and did not indicate that every depicted source was integrated or available.
Source groups, counts, and individual selection controls make the proposed research inputs available for inspection.
03. Making AI Research Inspectable
Keep the evidence accessible before synthesis
Deep research can produce a substantial set of potentially relevant findings. Researchers need a way to examine that material and determine what belongs in the analysis.
I placed evidence review between extraction and report generation. The design supported reviewing, filtering, sorting, and selecting findings, while maintaining access to source information, publication details, and links.
Compact and expanded views supported different review levels. The compact view emphasized titles and source metadata for scanning, while the expanded view included key takeaways and additional details. Filters narrowed results by selection status, publication year, and source.
Together, these controls made evidence selection an explicit part of the workflow, placing researcher judgment between AI-assisted discovery and final synthesis.
Researchers could scan results, inspect supporting details, and select findings for the report while retaining access to the original sources.
DESIGNING FOR TRUST THROUGH INSPECTABLE EVIDENCE
Design contribution: Make the basis of the research visible at the points where researchers exercise judgment.
Question refinement exposed ambiguity, source selection clarified inputs, and evidence review maintained the connection between findings and source material. These interactions supported the brief’s requirement for sourced, auditable outputs and extended that intent into the report design.
Source details remain alongside the finding before it contributes to synthesis.
04. Turning Research Into a Usable Deliverable
Create a structured report that supports continued analysis
The final stage organized selected evidence into a research report intended for further review, communication, and use.
I structured the output with an executive summary, sections, supporting tables, inline citations, and a reference list. Section navigation enabled movement through longer analyses. Export and modification controls allowed users to take the output beyond the workspace or create new versions.
Completed research steps remained visible above the report. Their summaries preserved context about the refined question, selected sources, and extracted evidence.
This unified the workflow in a persistent deliverable, with supporting references available for inspection alongside the synthesized findings.
The proposed report combines structured findings, references, navigation, and controls for export and modification. Report content is illustrative.
Designing for the Broader BioCentury Ecosystem
Define interaction patterns that extend beyond Bernie
I positioned Bernie within BioCentury’s broader working environment. The entry wireframe connected research initiation to site navigation, recent projects, report preferences, and audience controls. Within the research workspace, project naming and library access gave individual inquiries a place within ongoing work.
The design also established reusable patterns for question refinement, source selection, progress visibility, evidence review, and report output. These provided a foundation for future AI capabilities across the platform while maintaining a consistent focus on researcher control and source visibility.
The intended contribution extended beyond a single research flow to an experience model that could inform BioCentury’s evolving intelligence platform.
The proposed entry experience connects Bernie to platform navigation, research preferences, and recent projects.
What the Work Established
A defined research architecture supported by detailed interaction wireframes
I translated BioCentury’s AI product direction into an experience architecture and wireframes spanning research initiation, question refinement, source selection, evidence review, and report generation.
The work made researcher control concrete through specific interactions: clarifying the inquiry, choosing inputs, inspecting findings, and reviewing a report with supporting references. It also defined how progress and completed decisions could remain visible throughout the experience.
The delivered architecture and interaction design provided a foundation for developing the proposed research assistant and exploring related AI capabilities across BioCentury. This contribution is the design itself; this work did not measure production adoption or improvements in research efficiency or commercial performance.