When I joined this project, the bank had a clear challenge: customer support agents were spending too much time searching through a massive internal knowledge base, while clients expected quick, accurate, and compliant answers. We saw an opportunity to use Large Language Models (LLMs) to bring real-time intelligence directly into the support experience—but the solution had to balance speed, trust, and strict financial regulations.

​

Framing the Problem

Agents worked inside a CRM they knew well, but every time a customer asked a nuanced question, they had to break flow—jumping to a separate knowledge system, interpreting results, and then rephrasing the answer in brand voice. This slowed response times, introduced compliance risk, and made training new hires harder. My task was to design an AI-assisted interface that solved those pain points without disrupting existing workflows.

​

Designing the AI Experience

I partnered closely with data scientists to refine prompt engineering and context modeling. Together, we tuned the system so suggestions weren’t just technically correct, but also factually reliable, on-brand, and regulation-safe.

From a UX perspective, my focus was on transparent interaction design. Every AI-generated suggestion came with clear visual indicators and links to the exact knowledge base articles used. Agents could see why the AI suggested a response not just the answer itself.

To build trust, I added a confidence-level meter and a quick fallback to manual search. Agents could always verify or override the AI, ensuring they never felt locked in by the system.

​

Seamless Workflow Integration

One of my guiding principles was: no extra clicks, no context-switching. The AI suggestions appeared inline within the CRM interface agents already used daily. This minimized training overhead and let the tool feel like a natural extension of their existing workflow rather than a new platform to learn.

​

Building for Everyone

Because this tool was mission-critical, accessibility wasn’t an afterthought. I designed the interface to meet WCAG 2.1 AA standards—with strong contrast ratios, full keyboard navigation, and screen reader support—so that every agent, regardless of ability, could confidently rely on the system.

​

Impact

The end result was a support interface that gave agents instant, compliant, and trustworthy answer suggestions without breaking stride. By embedding AI directly into their flow of work, we reduced response times, improved first-call resolution, and made the support experience both more efficient and more human.