We collaborated with NVIDIA engineers Jessica Jiao and Anbang Xu to design concepts for an enterprise RAG AI chatbot that helps investors understand public financial data sources.
I conducted generative and evaluative research, deriving insights that led us from problem discovery to final design. I also took the lead on creating paper and digital prototypes for user testing, crafting our final interactive experience.
fiNVIDIA was positively received by key stakeholders, with our designs added as the next step in the product roadmap.
Click for prototype
fiNvidia
Empowering independent investors with Generative AI
TEAM
ROLE
DURATION
Anbang XU
Bridget Ho
Jessica Jiao
Neha Choudhary
Mar 2024 -
Dec 2024
Product Designer
CONTEXT
PROBLEM SPACE
Many people believe that investing is too complicated or risky – in fact, 48% of Americans don’t own any investment assets. NVIDIA saw the need for finance-focused AI and created an LLM that breaks down reports into actionable insights.
Scout Bot was an MVP enterprise concept that prioritized functionality over innovation. It only supported text-based input and output, though data visualizations had been requested to reduce investment research time.
It was critical that we understand how investors use public information to make decisions. We spoke to 4 expert investors and 2 beginner investors, discovering that research priorities differ based on experience level.
Sally, Expert Investor
"I want to quickly analyze what's relevant so I can decide whether to continue investing or not."
Matthew, Beginner Investor
"I want to understand what's important and what these updates mean so I can invest confidently."
PRIMARY RESEARCH
Defining investors' top needs and motivations
DIRECT SOURCES FEEL CREDIBLE
Earnings conference calls, SEC reports, and news articles that come from the company are viewed as truthful and unscripted.
Identifying trusted sources
SKEPTICAL ABOUT AI
Investors worried about accuracy, reliability, and predictive ability. Even if they use AI, they double-check answers.
We knew that different designs would appeal to different investors, so the first few sketch iterations included dashboards, document chats, and portfolio suggestions. We spoke with Jessica and Anbang to strategize an MVP plan and prioritized the dashboard and AI document chat as it aligned better with the LLM.
IDEATION
Exploring AI use cases with engineers
Collaborative design with investors
We wanted to bring investors into the design process through a participatory workshop. I developed a paper prototype that we tested with 6 peers to validate the concepts and initial user experience.
AI document chat
AI document chat
We updated the mid-fidelity Figma prototype based on the feedback, focusing on AI discoverability, using plain language to improve understandability, and developing a robust citation feature.
Dashboard with embedded AI
Chat view with pop-up citations
We tested key interactions like finding citations, using the AI chat, and visualizing data on the dashboard. Through 9 usability tests, we pinpointed 3 major usability issues that we addressed through design iterations.
design decisions
Evaluating & improving the user experience
Decision 1: Consistent sidebar affordances
Inconsistent component designs led 6 of 9 participants to struggle with identifying clickable elements. To address this issue, we redesigned the UI with clear labels, added dropdown arrows, and standardized primary and secondary buttons to highlight interactive elements.
Decision 2: Shortening citations for clarity
Although participants easily found the citation pop-up, 5 out of 9 struggled to find and understand the source document link. We refined the design by experimenting with layouts and text length, landing on a shorter preview and descriptive button that prompts users to open a split-screen view.
Decision 3: Guiding investors to the AI chat
There was a 50% misclick rate when looking for the embedded AI chat on our dashboard. We redesigned our flow to emphasize the "Create with AI" feature, added suggested actions in the embedded chat, and included multiple entry points to the AI feature to improve discoverability.
solution
Your personalized financial AI Assistant
fiNVIDIA is a multimodal AI tool that combines Generative AI and Retrieval-Augmented Generation (RAG) to transform public financial data into trustworthy, actionable insights.
Visualize trends with AI
The embedded AI chat helps users visualize past and predicted trends with natural language, so they can identify hidden trends to make strategic decisions.
Synthesize your documents
Users can upload financial documents and customize their knowledge base. fiNVIDIA will search across sources to answer questions, becoming a personalized expert in the documents that matter most to investors.
Sentence-level sentiment analysis
Color-coded highlights guide attention towards positive and negative statements in financial documents, so investors can focus on the important points.
Trustworthy AI insights
To build credibility in AI-generated responses, every answer and visualization links to its original data source. With RAG citations, users see exactly where each answer comes from.
User testing with AI functionality
One limitation was the lack of AI interactions. I would conduct usability tests with the interactive MVP to improve the product experience based on customer feedback.
next steps
Proposed future work
Enterprise-focused testing
We primarily tested with independent investors because of recruiting limitations. I would love to conduct user testing with NVIDIA's Scout Bot for financial earnings questions.
Jessica's engineering perspective greatly streamlined our process from ideation to final design hand-off. As a designer, I learned to lead effective collaboration by presenting early concepts, testing ideas with an open mind, and balancing between customer needs and business goals.
We presented our designs to the internal NVIDIA team in January 2025, and the concept was well-received. It fulfilled Anbang's vision, aligned with engineering goals, and served as an ideal next step for Scout Bot once they iron out the LLM's reliability.
reflections
Complex documents, clear insights
