Recommendations for the Library of Congress
My role
As the team lead for the University of Michigan School of Information Alternative Spring Break trip, I…
Led 9 informatics students
Spearheaded roadmaps & project board, task assignments, standups, troubleshooting & stakeholder communication
Conducted UX research
Developed personas
Improved IA
Prototyped a live GitHub site with refined design system (Jekyll vs HTML)
Presented findings to five Library of Congress Labs employees
Scope
UX Research (Primary & Secondary)
Accessibility & Site Audit
Recommendation Slide Deck
Prototype Design Solutions
Tools
Google Suites, Figma, GitHub, WAVE, aXe
Duration
On-Site: March 3rd-7th, 2025
Preparation Period: January-March 2025
Client
Library of Congress Digital Initiatives Division (otherwise known as LC Labs)
Skip to:
Brief
Expectations
As a team, we initially thought…
The focus would primarily be on improving accessibility in LC Labs’ GitHub Pages. This entails meeting WCAG or other standards
The project would require less research and minimal direct contact with stakeholders.
The main goal was to refine overall usability.
Pivoting focus
After talking to the client in person on-site…
We realized the scope of the project would be larger due to the client’s needs: recommendations regarding overall usability, accessibility, and content on the open source AI Planning Framework.
The expected final deliverable would be a slide deck presented at the end of the week to the LC Labs internal team.
Background
AI is reshaping access to library collections, making clear and ethical documentation of its use essential. Inconsistent practices can create challenges when generating, interpreting and maintaining information. This project focuses on refining LC Labs’ GitHub Pages to improve usability, accessibility, and clarity— ensuring AI transformations are well-documented and easily understood.
Current GitHub Page Embed
Problem statement
How might we improve the usability, accessibility, and clarity of the AI Planning Framework so that users can easily navigate, understand, and apply responsible AI practices?
Constraints
To highlight opportunities for future improvements, we want to acknowledge the scope of our process. While we gained valuable insights from internal users, incorporating feedback from external users—who represent a larger portion of the target audience—would further enrich our findings. With additional input, our conclusions could evolve to better reflect the needs of a broader user base. Given our project’s limited timeframe, we conducted two interviews, which provided meaningful insights but also present opportunities for deeper exploration in future iterations.
Research
Primary
We interviewed two readily available participants who were familiar with the AI Planning Framework via Google Meets in order to understand user sentiments and perceptions regarding their experience using the GitHub page.
Findings from the interviews were:
Difficult to understand terminology
Unclear value
Unclear user
File format disrupted browser experience
I would probably leave the site if I didn’t already know what it was.
I wouldn’t understand what IDIQ means if I didn't work for the Library… feels like an insider term.
Backed by research, we developed a persona representing a librarian who seeks ethical and effective guidance on using AI.
Background
With over 8 years of experience in public and academic libraries, Lucy has always been passionate about information literacy and emerging technologies. Recently, she has taken a keen interest in AI and LLMs to better understand their implications for knowledge management, digital archiving, and public information services.
Resourceful
Analytical
Curious
Detail oriented
Goals
Plan for upcoming AI experiment
Understand risks and benefits of AI
Needs
Credible resources
Accessible files
Time efficient experience
Information about AI Planning
Pain Points
Confusing website terminology
Hard to digest web content
Trouble navigating site
Frustrating external links
Secondary
Because we had limited interviews due to time constraints, we leaned heavily on secondary research. This included initial background research, comparative evaluation, and citation chaining to understand how existing AI frameworks outlined best practices for information hierarchy and what online resources provided in terms of usability and features.
Key insights from comparator analysis:
100% of comparators separate content into sections by topic
100% of comparators include a sitewide navigation bar
Only 1 comparator required non-browser software to view work aids
Site audit
We also conducted a site audit of LC Labs and GitHub Pages to ensure that WCAG standards were met leveraging tools such as aXe and WAVE.
Key insights from the site audits:
Non-responsive
Worksheets downloadable in docx format
Lack of discoverability
Lack of scannability
Citation chaining
Citation chaining enabled us to conduct an in-depth exploration of AI frameworks and best practices. By identifying key research and articles on responsible and ethical AI usage, we traced their cited sources as well as subsequent citations. This iterative process allowed us to uncover a broader network of insights, deepening our understanding of how AI frameworks are structured, documented, and implemented effectively.
Ultimately, citation chaining reinforced our findings from secondary research, aligning with our comparator analysis of best practices in presenting AI frameworks. While it didn’t introduce any unexpected insights, it was a valuable exercise that affirmed how AI frameworks are currently being presented across various websites.
Research synthesis process
Affinity mapping
After conducting research, we organized data in a spreadsheet labeling each with a unique identifier and affinity mapped as a team.
Affinity mapping allowed us to identify the following key clusters:
Card sorting
To address terminology confusion, we conducted unstructured card sorting to group terms into logical categories. This helped inform a more structured approach to content clarification such as analyzing which terms could exist as a popover or what terms should be clarified in a glossary located on a separate page.
Key Findings
Through stakeholder interviews, comparative analysis, and research synthesis, we identified the following challenges:
Unclear Target Audience & Purpose – The intended users of the framework were not well-defined, leading to confusion about the page’s objectives.
Navigation Issues – The single-page markdown file lacked a structured navigation system, making it difficult for users to find relevant content.
Terminology Barriers – Internal and technical jargon made the content difficult to understand for users unfamiliar with LC Labs’ specific terminology.
Content Accessibility – The worksheets were only available as .docx downloads, which disrupted the user experience and assumed all users had access to Word or Pages.
Lack of External References – There was no glossary or external reference guide, leaving users without additional context for key concepts.
Gaps & opportunities
From our findings, we began to conceptualize opportunities and gaps.
Gaps:
1. Navigation: The current GitHub Page lacks navigation & connectivity to the main site and other related resources.
2. Content Format: Current worksheets are only downloadable in docx files, assuming users have access to Word or Pages on their device, also disrupting browser experience.
3. Overview: Current page does not include an overview for the page content, assuming all users know why & how to use it.
4. References: The page also does not include an external reference guide with the bulk of the information.
With those gaps in mind, we conceptualized opportunities for LC Labs to be able to improve the AI Framework.
For navigation, adding this page onto the existing website would make the page more discoverable and credible. Also allowing the opportunity to link out to other related resources on the website like a glossary or references page can help users to understand the content they are reading.
For Content Format, it may be beneficial to convert worksheets/files to downloadable pdfs, interactive forms, or browser pages with data privacy disclosure to remove assumptions and address content accessibility. This will also allow users to stay within the browser and increase user retention.
Regarding the idea of an overview, including an About Page to provide context about who the framework is for can provide clarity and efficiency for the content and the user.
Lastly, include reference links throughout the site to inform users on a deeper level about certain terms, elements, and implementations that they might not gather just from the main page can help users to have a better understanding of what they are reading.
Prioritization
Then, we used a feature prioritization matrix in order to determine what to prioritize in our roadmap.
Ideation
After identifying our priorities, we began to ideate how we wanted to draft our recommendations and findings. Although the project brief only required a slide deck deliverable, we believed that it would be beneficial to develop an interactive live prototype for the following reasons:
1. LC Labs primarily uses GitHub Pages internally meaning that a static design file (Figma) being handed-off would not integrate seamlessly with the team’s current internal processes.
2. Our client communicated a nice-to-have prototype deliverable but as a team, we wanted to push ourselves and ensure that our work would be helpful and most importantly, useful.
Additionally, we observed that the LC Labs team relied on Jekyll markdown for their deployed GitHub Page. While we recognized the advantages of Jekyll—such as standardized templates and ease of use for non-technical users—we also wanted to evaluate its limitations. To gain clarity, we consulted with LC Labs about their use of Jekyll and, after confirming that we were not restricted to it for the final prototype, proceeded with an HTML-based live prototype hosted on GitHub.
We also spoke with an Innovation Specialist on the LC Labs team (who leads communications, knowledge management and website updates for Labs) to have a better understanding of the LC Labs goals and roadmap for ensuring that recommendations were tailored to the needs of the LC Labs team.
Solution
Finally, our recommendations were outlined as the following:
Responsive design
Over 40% of users visit LoC websites on a mobile device, making a responsive design essential for accessibility and user experience.
Search engines prioritize mobile-friendly websites, increasing visibility and discoverability.
Discoverability
Based on Adobe Analytics, the majority of users access LoC websites through search engines, with over 80% arriving via Google search. This makes SEO optimization essential for enhancing discoverability.
In the last fiscal year, over 80% of users
navigated to LoC websites
using search engines
To improve visibility, the webpage could be integrated into the main LoC Labs website, strengthening navigation and domain authority. Additionally, both interviewees mentioned accessing the site either through direct URL sharing from colleagues or precise Google searches.
Strategically incorporating keyword variations—such as “AI Planning Framework”, “AI in libraries”, and “AI and digital preservation”—throughout titles, headings, meta descriptions, and body text can further enhance search engine ranking.
Navigation
The current AI framework at LC page is linked a few paragraphs down from the top of the page. This is an issue because:
We are unable to locate the planning framework as an obvious option from the navigation
As found in our user interviews, many of the flaws such as the lack of external references, information organization, a navigation bar on the side would be very helpful
The planning framework link is buried in "AI at LC" on the Labs page.
Navigation bar
A navigation head would provide a clearer access point for the framework, ensuring consistency with the headers used across LoC Labs pages.
Navigation head would allow for clearer access point for the framework
Side navigation
A side navigation menu would help separate content clearly, allowing users to focus only on relevant information. It would also enable users to scan for key content more efficiently. The structure and format would be inspired by other Library pages for consistency and usability.
Enable users to scan for relevant content & structure and format inspired by other Library pages (integrates with current design system)
File format
Interviewees expressed surprise that the worksheets were provided in DOCX format rather than PDF, as this disrupted their browsing experience. From the team's perspective, while the current worksheets are in DOCX format, PDFs are generally preferred for better usability within web browsers. However, this assumes that PDFs can be made more accessible; if not, the accessibility workflow effort may be a tradeoff for using PDFs.
Current worksheets take users out of browser due to docx format
Currently, GitHub Pages (and Jekyll) could be retained to allow authors to create worksheets without requiring technical knowledge of HTML. However, for a better user experience, worksheets could be implemented in HTML, either statically or interactively. The tradeoff with interactive forms is that they require greater technical expertise to implement and may introduce data privacy concerns.
Terminology
Many terms on the page, particularly in the “Experiment” section, may be unfamiliar to users. These include “IDIQ”, “Model Card”, “Data Cover Sheet”, “Curatorial Provenance”, “NLP”, “Baseline”, and “Benchmarking”.
From this analysis, we concluded that additional card sorting and research are needed to better understand the target user base’s familiarity with these terms. To improve clarity, concepts and terminology could be explained through accessible methods, such as tooltips providing definitions directly on the page or a dedicated glossary page for reference.
Exploring the tradeoffs between glossary page and popovers in the form of tooltips
We recommend conducting further research, including A/B testing and additional card sorting, to determine which terms should be presented via tooltips versus included in a separate glossary page.
Accessibility
The LoC Labs web page currently lacks a "Skip to Main" link, which is essential for keyboard users to bypass site navigation and access content more efficiently. Implementing this feature would enhance accessibility and ensure better compliance with accessibility standards.
First tab on page goes straight to navigation
Bugs
The AI at LC page currently marks both "Work" and "AI at LC" as selected in the header. This may stem from the assumption that users navigate to the page through Work > Experiments. However, this could cause confusion and may need to be reviewed for clarity in navigation.
Incorrect State
Final Prototype
Client feedback
We finished our time at the Library of Congress by presenting our findings over Microsoft Teams as LC Labs team members were located all over the country.
Overall, the feedback we received was highly positive. Some members of the LC Labs team were particularly interested in the card sorting results, noting that defining terms had been an intensive process—for example, it previously took six months to define AI. Given this, they were eager to learn more about how we approached term classification and organization.
The live prototype was well received, with team members expressing enthusiasm, especially given the tight time constraints under which it was developed. Many found it unexpected yet valuable.
To ensure a smooth handoff, we provided LC Labs with our complete work process and final deliverables, including a two-minute prototype demo video with closed captioning.
Reflection
Learnings
One of our key takeaways was gaining a deeper understanding of AI frameworks and their value for practitioners. Through our research and discussions, we realized that implementing AI is far more complex than we initially thought, reinforcing the importance of being informed users of AI technologies.
We also learned a great deal about LC Labs and the impactful work being done—from experiments and research to AI initiatives. Exploring how these efforts align with the mission of the Library of Congress provided valuable context for our work.
Additionally, we had to navigate constraints effectively to maximize our project goals within a limited timeframe. This challenge pushed us to think critically about prioritization and efficiency.
Finally, understanding internal processes was crucial in shaping our recommendations. By tailoring our work to fit within existing workflows, we aimed to ensure that our contributions would be both practical and beneficial to the LC Labs team.
What I'd do differently
Given our primary constraint of time, there are several things we would do differently if we had more time.
First, we would prioritize conducting at least five in-depth user interviews with target users, specifically practitioners. This would help us better understand their needs, challenges, and expectations, leading to more user-centered insights. Additionally, we would schedule these interviews early in the process to ensure that research is prioritized from day one, allowing us to make more informed decisions throughout the project.
We would also conduct a deeper analysis of file formats. While we recognize the accessibility benefits of DOCX files and the potential user retention advantages of PDFs, we were not entirely certain about the full range of trade-offs between them. Conducting an accessibility analysis comparing DOCX, PDFs, and other alternatives—such as an online assessment tool—would provide a clearer understanding of the best format for both usability and accessibility. This deeper exploration would enable us to make more concrete and well-supported recommendations in the future.