Evolving financial search

Summary

As AI reshapes digital experiences, Fidelity Investments recognized the limitations of its legacy search interfaces: rigid, page-bound workflows that slowed users down. This project represents the beginning of our exploration, reimagining search as an AI-driven, in-app conversational assistant. By breaking down the walls of traditional prototyping with tools like Cursor, I began demonstrating how UX teams can rapidly craft high-fidelity, test-ready experiences, moving beyond Figma and into a new era of experimentation.

Problem

The expectations for search have shifted dramatically. Tools like Perplexity, Claude, and ChatGPT have trained users to expect fast, contextual, and conversational experiences that feel less like searching and more like collaborating. In contrast, Fidelity's existing search required users to leave their current page, disrupting flow and forcing them into a narrow, one-and-done interaction model. It lacked memory, depth, and adaptability. This presented a clear opportunity: to rethink search as an embedded, AI-powered assistant that understands follow-ups, stays within context, and evolves with the user.

Objectives

• Prototype a conversational search experience within existing Fidelity apps.
• Enable seamless follow-up queries without losing page context.
• Showcase rapid prototype creation and testing using AI-first tools.

Research & Initial Insights

To inform our direction, the team and I pursued three complementary streams of research that helped validate our concept and identify key design opportunities early on:

• Stakeholder Workshops: We facilitated live demos and collaborative workshops with product, engineering, and compliance teams. These sessions gave stakeholders a chance to experience the potential of a conversational search assistant firsthand. Their feedback helped shape the scope of our MVP and align technical feasibility with user value.
• Competitive Scan: We studied modern AI-first tools like Perplexity, Claude, and Magnefi to understand how they handled disambiguation, guided flows, and contextual follow-ups. Features like context chips, inline suggestions, and persistent conversational memory surfaced as core expectations for today's users and set a high bar for interaction quality.
• Usability Triaging: We conducted informal hallway testing with early prototypes built in Cursor. These tests surfaced patterns around hesitation with input phrasing, lack of clarity around the assistant's capabilities, and the need for visible state cues. This helped us improve both the UX writing and UI responsiveness in subsequent versions.

Rapid Prototyping Cycle

I pivoted from static design files to AI-enabled prototype builds:

• v1 – Solo Proof-of-Concept: Built in Cursor within hours: chat interface, search arrays, contextual interactions, and voice input. Demonstrated to stakeholders how quickly ideas can become clickable reality.
• v2 – Team Structure & Flow: Collaborated with UX peers to define a modular template system and component library. Held weekly critiques to align on entry points and interaction patterns.
• v3 – User-Focused Demo: Delivered high-fidelity end-to-end scenarios (e.g., ticker lookup → transaction drill-down) in under a week. Prepared to launch early user tests, ongoing at the time of writing.

Prototype Evangelism

Changing the UX Playbook

• 'Figma is Dead' Mindset: AI-driven tools like Cursor let us bypass traditional wireframes and jump straight into building functional, high-fidelity prototypes that feel real and testable from day one. This not only saves time but also aligns better with how today's teams evaluate and iterate on ideas.
• Stakeholder Buy-In: By demoing live, interactive prototypes early in the process, we transformed stakeholder attitudes from passive observers to engaged collaborators. These demos helped collapse decision timelines and build confidence around bold design directions.
• UX's Evolving Role: With AI increasingly automating routine UI generation, the role of UX is shifting toward systems thinking, interaction modeling, and rapid experimentation. Our value lies in turning ambiguous problems into functional, testable experiences quickly, making prototyping fluency a critical skill in this new era.

Architecture & Key Components

Scalable, Modular Design

Our architecture was intentionally built with extensibility and collaboration in mind. We chose Next.js for its routing and API flexibility, paired with Tailwind CSS and shadcn-ui to ensure design consistency and fast front-end iteration. Every component was crafted to be both expressive and adaptable, allowing us to evolve with minimal overhead as new requirements emerged.

• DialogueArea: Acts as the conversational orchestrator, managing state, determining user intent, and rendering the appropriate response templates dynamically.
• EnhancedInput: A custom-built, multi-modal input system that supports:
  • Alias Support for flexible input recognition
  • Voice transcription for hands-free interactions
  • Context chips to disambiguate queries and enable agent-specific querying
  • Document upload for direct PDF or statement queries
• Template System: Three tiers of templates (Large, Medium, Small) that scale with query specificity.
• Card Library:
  • Overview Cards: High-level context or summaries
  • Data Cards: Detailed financial metrics and visualizations
  • Action Cards: Interactive prompts for user actions

Next Steps & Early Plans

• User Testing (Ongoing): Running moderated sessions (n=12) to gather qualitative feedback on interaction patterns, language tone, and component behavior.
• KPI Definition: Establishing metrics like time-to-feedback, task completion rates, and ability to sustain context across multi-step queries.
• A/B Testing Roadmap: Comparing user experience between conversational and traditional search flows, focusing on efficiency, trust, and perceived intelligence.
• Live Data Integration: Beginning to connect real-time stock market APIs to simulate realistic query responses and dynamic content.
• Design System Alignment: Refining visuals to align with Fidelity's design system, improving consistency, accessibility, and developer handoff readiness.

Stumbling Points & Lessons Learned

Stretching Beyond Comfort Zone

Building this semi-complex app was a true stretch assignment that pushed me beyond my front-end development comfort zone. Along the way, I wrestled with event handlers, structured data arrays, and logic flows inside a modular app environment. These challenges became growth opportunities.

Leveling Up Next.js Architecture

Through the process, I gained a stronger grasp of how to structure scalable applications in Next.js, especially the interplay between routing, component composition, and shared utility logic. My ability to build smart, testable layouts improved dramatically.

Mastering AI-Native Workflows

Equally important was improving my use of Cursor. I learned how to reverse-engineer prompts when I hit dead ends and use memory traces to get back on track faster. I became better not just at writing effective prompts but at designing prompt chains that mirrored real workflows.

The Importance of Memory & Context

The biggest turning point was realizing that managing chat context wasn't just helpful, it was essential. As prototypes grew, I needed a strategy to maintain coherence across long sessions. I designed a structured memory bank and committed to a schedule of updates, based on the Cline memory protocol (with a few of my own tweaks). This allowed me to carry nuanced context forward, accelerate iteration, and keep the cognitive thread intact. Without it, progress would have been much slower.

Reflection

The Fidelity Conversational Search initiative reinforces a powerful truth: AI-first prototyping is not just a trend, it's a shift in how UX work gets done. Through this project, I experienced firsthand how stretching beyond traditional toolsets unlocks speed, depth, and creative momentum. Building and refining a semi-complex app pushed my front-end fluency forward, while memory design taught me the critical importance of managing context at scale. By pairing high-fidelity prototypes with real-time feedback loops and emerging AI workflows, we didn't just accelerate iteration, we transformed how quickly ideas could be tested, trusted, and evolved across teams. It's an approach that turns UX into a driving force for innovation.