AI in Quality

Overview

I led the end-to-end design of 5+ compliance processes for an AI-powered enterprise product that transformed how the MedTech sector manages quality management. The product empowers quality and compliance teams to investigate and audit complaints efficiently, dramatically reducing manual effort and operational costs while increasing efficiency.

Impact at a Glance

• 60% reduction in manual documentation effort

• 38% reduction in errors when copying and pasting information into the records

• 400+ implementation issues identified through Design QA and ensured they were resolved before the release

• $62M true cost savings projected by 2028

The Challenge

The organization's Quality Management System was fragmented across 30+ technology applications with varying levels of integration. These inefficiencies cost the business $3.1M annually due to manual processes, including employee training that consumed 500+ hours every year across the MedTech sector. The product aimed to significantly reduce manual effort while providing a single platform to manage all quality processes.

Quality and compliance professionals faced significant challenges:

• Complex, manual workflows requiring employees to navigate disconnected systems

• Fragmented data creating duplicative work and inconsistent records

• High rejection rates forcing users to rework entire records and manually find errors

• Significant time spent on transactional tasks rather than high-value activities

Navigating Constraints as a Designer

Compressed timeline: I joined two months after product goals were set and the technical framework was locked.

Technical limitations: The framework restricted certain interaction patterns. When designing how users edit AI-generated responses, my ideal approach wasn't feasible. I had to be creative to find other solutions that aligned with the users and the stakeholders.

No Figma dev access: Without dev mode, early builds diverged from design intent. I created a mini component storybook with CSS specs, held regular working sessions with developers, and tested directly in the dev environment to catch issues before they shipped.

Design Solution

The goal was to significantly reduce manual documentation using AI assistance, help quality professionals reduce errors (which cause records to be reopened), provide a single place to edit and manage records, and support auditing of the issue and the corrective steps taken to resolve the complaint. The interface needed to feel intuitive from day one and scale to support power users, with a strong emphasis on learnability.

From discovery research, two insights shaped the direction of the design:

• Quality professionals were managing 50+ workflows for a single record

• Every piece of the complaint and corrective processes documentation was written manually.

• 
  

Key design decisions:

• Flexible chat panel overlays the workspace so users can review records while AI processes requests. A full-screen version supports documentation work for global users without internal app access.

• Smart tab system unifies Current Chat, Chat History, and Notifications, letting users run reports, revisit chats, and receive completion alerts. It addresses the research-identified need to multitask during lengthy AI report generation.

• Tiered guidance system offers guided prompts for new users and a smart writing panel for experts to refine drafts with AI.

• Intelligent information aggregation uses AI to combine data from records and connected quality systems, eliminating manual copy-and-paste and reducing human error by 38%. Visible provenance for each data point keeps humans in the loop.

Design QA

I volunteered to lead Design QA to test every feature in the development environment, and it became essential to my work. Throughout the design process, I conducted heuristic evaluations and WCAG 2.2 accessibility audits, and I tested the process logic and AI output quality with stakeholders to ensure we produced high-quality outputs in compliance with the procedure.

This effort also identified 400+ implementation issues, and I partnered with the development team to ensure we fixed them all before launch. This effort also significantly reduced bugs reported by the testing team, saving many story points.

Results and Impact

What I Learned

This project taught me that design excellence isn't about ideal conditions; it's about making smart decisions within real constraints.

• Constraints force creativity: The technical limitations that initially felt like obstacles- the locked framework and the missing Figma dev seats pushed me toward solutions I wouldn't have considered otherwise. The component storybook I created as a workaround became a reference that outlasted the original problem.

• Process ownership is design work: The handoff process, QA checkpoints, and developer communication cadence I built weren't distractions from design; they were the reason impactful design reached production.

• Dual-path interfaces beat one-size-fits-all: The Guided/Draft mode worked because we stopped trying to find a middle ground that would partially serve everyone. Designing explicitly for different expertise levels delivered better outcomes for both groups than a compromised single experience ever could.

​​​​​

Confidential business details have been generalized for portfolio presentation