A leading semiconductor manufacturer faced critical operational inefficiencies that were directly impacting their bottom line. Their workforce was polarized: seasoned technicians held deep tribal knowledge but rarely documented it, while junior technicians constantly needed guidance to resolve equipment errors.

 

The Business Impact:

• Revenue loss from extended equipment downtime during critical manufacturing

• Inefficient resource allocation with senior technicians constantly interrupted

• Knowledge silos preventing organizational learning

• Scalability challenges preventing workforce expansion

 

The Root Problem: The company had no centralized system for technical knowledge. Critical information was scattered across equipment manuals, technician logs, chat logs, and undocumented expertise. When complex, rare error codes appeared, junior technicians faced significant delays waiting for senior technician guidance.

Research Strategy

As Senior Design Lead, I designed a comprehensive research approach to understand both the technical and human challenges:

Remote Discovery Workshops

• Conducted virtual screen-sharing sessions with technicians and managers

• Observed their painfully slow, outdated systems firsthand

• Watched them resort to Teams chat for real-time problem-solving

• Led virtual discovery workshops to understand their existing workflows

Key Research Insights:

• Senior technicians were frustrated by constant interruptions and poor documentation processes

• Junior technicians spent excessive time searching disconnected systems

Critical finding:

• Not everyone wanted a chatbot interface—older, less tech-savvy senior technicians preferred traditional click-through methods with visual data access

Surprising discovery:

• Teams chat had become an informal knowledge base where senior technicians repeatedly answered the same questions

 

User Understanding

I created distinct personas and journey maps in Mural to capture the different needs:

• Senior Technicians: Pattern recognition experts who needed quick, visual access to comprehensive data but weren't comfortable with conversational interfaces

• Junior Technicians: Guidance-seekers who were familiar with chatbots and benefited from conversational, step-by-step support

This insight fundamentally shaped my design approach—rather than forcing one interaction model, I would create flexibility within a single interface.

Design Leadership & Collaboration

As design lead, I mentored a junior designer through the entire process, teaching Figma tools and facilitating collaborative prototyping sessions. I worked closely with the AI engineer to understand technical constraints, ensure the solution being built was plausible, and define realistic system behaviors.

​

Solution Architecture

The Core Insight: Instead of separate interfaces, I designed one unified system with dual interaction modes—a smart search hub with an optional chatbot feature that users could toggle into when needed.

Interface Design:

• Primary Interface: Traditional search and navigation for quick access to all available data

• Optional Chatbot Feature: Conversational interface that users could access when they needed guided support

• Flexible interaction: Users could switch between modes based on their comfort level and specific needs

 

Overcoming Design Constraints

Working with significant limitations, I had to make strategic design decisions:

• Limited system access: Only had access to a few screens shared during discovery workshops

• Brand constraints: No access to brand guidelines, so I extrapolated design language from their existing website

• System integration assumptions: Had to make assumptions about how alerts would be surfaced and validated these with stakeholders

 

AI Behavior Design

I collaborated with the AI engineer to define critical system behaviors:

• Confidence indicators: Clear visual cues when AI certainty was low

• Source citation: System always attributed information to specific ingested documents

• Honest limitations: System explicitly stated when it didn't know rather than making up information

• Quality focus: Trained AI to only report on data it had actually ingested

 

Testing & Validation Strategy

I established a rigorous testing approach:

• Answer key methodology: Customer and technicians provided acceptable answers for test scenarios

• 90% accuracy threshold: Trained the AI until it achieved 90% accuracy against the answer key

• User testing: Conducted online user testing with concept prototypes with actual technicians

 

Product Strategy & MVP Definition

I defined a clear feature prioritization strategy:

• MVP focus: Smart glossary functionality using GenAI to ingest documentation and identify error types with potential resolutions

• Phased data integration: Started with high-quality sources (manuals, video walkthroughs, well-documented resolutions) before including chat logs

• Future phases: Real-time system integration to automatically surface error codes and types

• Key metrics: Time to identify errors was identified as the most critical measurement, especially for complex and rare error codes

 

Implementation Strategy

The solution addressed a larger organizational documentation problem through staged implementation:

• Phase 1: Ingest existing documentation and create smart search capabilities

• Phase 2: Add chat logs and less structured data sources

• Built-in validation: Automatic logging of AI suggestions with periodic check-ins to close out error events

• Continuous improvement: Data quality checks and technician validation built into the workflow

Technical Achievement

The AI system achieved a 90% accuracy rate in solution recommendations when tested against the technician answer key through data training. We established automated knowledge capture through system logging of AI suggestions with periodic validation requests to build ongoing documentation. The solution included confidence indicators and source citation features that helped build user trust in the AI recommendations.

 

User Adoption Success

The flexible interface approach successfully accommodated users across different comfort levels with technology. Senior technicians embraced the visual search functionality while junior technicians leveraged the optional chatbot feature. This dual approach reduced the constant interruption burden on senior technicians while empowering junior technicians to work more independently.

 

Business Expansion

The client was impressed with the solution and sought to implement it enterprise-wide. They introduced our team to different business units, opening up possibilities to scale the solution across the organization. The project began addressing the broader documentation and knowledge management challenges that extended beyond just the technician workflow.

 

Design Process Success

I successfully presented the dual-interface insight to stakeholders and gained approval for the flexible approach rather than a single interaction model. The effective partnership with the AI engineer resulted in a realistic, implementable solution that balanced user needs with technical constraints. Through this complex AI design project, I was able to successfully mentor the junior designer and demonstrate effective design leadership.

This project demonstrated the critical importance of user-centered AI design. Rather than imposing a single interaction model, I created flexibility that accommodated different user preferences and technical comfort levels. The success came from understanding that the real challenge wasn't just technical—it was organizational and cultural.

Key Design Learnings:

• Flexibility over uniformity: Different users need different interaction paradigms, even within the same role

• AI transparency builds trust: Confidence indicators and source attribution were crucial for user adoption

• Phased implementation reduces risk: Starting with high-quality data sources established system credibility

• Design leadership requires technical collaboration: Working closely with AI engineers was essential for realistic solutions