The knowledge Intel needed most often lived in one person's head, and nowhere else.
Project
iGPT Trainer and Marketplace
Client
Intel
My Role
Lead UX/UI Designer and Researcher
Timeline
Results
1 year
First-of-its-kind AI training and marketplace platform, built at Intel with no existing system to model it on
Positive reception across every level of the organization, from non-technical marketers to senior AI engineers to executive stakeholders
Group sharing, a feature design and engineering proposed rather than were asked for, adopted into the final product roadmap after stakeholder review
The My Assistants dashboard, showing the range of AI tools already being built and shared across the organization.
The Challenge
At an organization the size of Intel, expertise was everywhere, but it was trapped. An engineer who understood a specific process, a marketer who knew a niche workflow inside out, that knowledge only moved as fast as someone's calendar allowed. Scheduling conflicts and departmental silos meant institutional knowledge slowed down the exact collaboration it was supposed to enable. Because AI-driven knowledge transfer was completely new territory for the company, there was no playbook to build from. We were designing the thing that would define what "sharing what you know" even meant at Intel going forward.
The group sharing step of the Publisher, letting a trained assistant be restricted to select teams instead of the entire company.
The Decision
Nobody asked us for group sharing. The original scope was simpler: train an assistant, publish it, let the company use it. But as we got deeper into the design, it became clear that "share everything with everyone" wasn't actually going to work. Some of the knowledge people wanted to train an AI on was confidential by nature, tied to a specific team or project, not something meant for the whole company to query. Design and engineering proposed a secure group sharing model as a solution: an assistant could be trained and published, but restricted to only the groups who should have access to it. When we presented the idea to stakeholders, they asked us to build it into the final product. It became one of the platform's core features, not because it was requested, but because we identified a real gap before it became a real problem.
The Assistant Trainer, built with contextual guidance and templates to help less technical users train an assistant with confidence.
Designing for Every Level of Expertise
What the Research Found
We assumed the terminology behind training an AI was simple enough to design around without much hand-holding. Research told us otherwise. Because generative AI was still new to most of the organization, the same interface that felt intuitive to an engineer could be completely opaque to a first-time user in marketing or operations. We couldn't design a static tool and expect it to work for every level of technical fluency. Instead, we had to design for teaching and learning alongside the product itself, building an experience that could onboard less technical users in the moment, and that would keep working long after our team had moved on to the next project.
Advanced search and filtering in the marketplace, designed to keep discovery fast as the number of shared assistants grew.
The platform had to work for two very different kinds of users at once. Someone training their first assistant needed guidance, not jargon. Someone searching a marketplace that would eventually hold hundreds of assistants needed precision, not a wall of options. For training, we built resources directly into the flow, including template starting points and contextual help, so a first-time user wasn't staring at an empty system prompt field with no direction. For discovery, we designed advanced search and filtering (by model, tags, author, update date) so that as the marketplace scaled, finding the right assistant stayed fast instead of becoming another kind of noise.
Marketplace listing states, showing public, private, and group-shared assistants side by side.
Intel engineers, developers, and cross-functional stakeholders, across AI design and training workflows, feasibility testing, user interviews, persona creation, and journey mapping to align the platform with real technical fluency across the organization.
Impact
Delivered Intel's first end-to-end AI training and marketplace platform, designed from the ground up with no prior internal benchmark to build from
Introduced group sharing as a new capability, later adopted as a standard part of the product after being proposed by design and engineering
Built an interface usable by both non-technical employees and senior AI engineers, validated through research across all levels of technical fluency
Received positive feedback from stakeholders at every level of the organization, from individual contributors to executives