Chegg Skills

The project transformed one course into a repeatable model for building many.

I led the evaluation of the MVP, defined the shared learning architecture, and designed reusable activity and content patterns. I also introduced AI-supported workflows that accelerated production while keeping learning strategy, accuracy, and final quality under human control.

Problem

Chegg wanted to build a portfolio of short-form technical courses for recent graduates, early-career professionals, and working adults developing in-demand skills.

The team had a working MVP. That solved the first problem: proving that a course could be built.

It did not solve the harder one.

Every new course would require different subject-matter expertise, examples, activities, and career outcomes. But if the team redesigned the structure and production process each time, the portfolio would become slow and expensive to build. If we standardized too much, the courses would lose the relevance and specificity that made them valuable.

The challenge was to scale the program without flattening the learner experience.

Research → Design

The MVP showed us what could work, but also where a one-off course would break under the pressure of scale.

I evaluated the existing experience across lesson structure, interaction patterns, workplace relevance, assessment, and production requirements. I used that analysis to separate the elements worth preserving from the decisions that needed a more systematic approach.

What the MVP revealed

Some lessons moved naturally from explanation to application. Others asked learners to absorb information without giving them a meaningful way to use it.

The production process presented a parallel problem. Without a shared architecture or review standards, every course would require the team to reconsider the same foundational questions:

• What should the learning sequence be?
• What kinds of practice should the course include?
• What did subject-matter experts need to provide?
• How would reviewers evaluate quality?
• Which decisions should remain consistent across the portfolio?

The MVP did not need to become a rigid template. It needed to become a flexible system.

Establishing design principles

I synthesized the evaluation into four principles that guided both the learner experience and the production model.

Every step should earn its place

Learners should understand what they are learning, why it matters, and how it moves them toward greater capability.

Application should drive the experience

Learners should repeatedly make decisions, solve realistic problems, and use new skills, not simply recall information.

Consistency should reduce friction, not flatten the content

Shared patterns should make courses easier to navigate while leaving room for the demands of different subjects and skills.

AI should support judgment, not replace it

Automation could accelerate repeatable work. Learning strategy, accuracy, context, accessibility, and quality still required human expertise.

These principles gave me a practical way to decide what to standardize and what needed to remain flexible.

They also provided shared criteria for evaluating decisions across instructional design, UX, subject-matter experts, and QA reviewers.

Design response: Give each course a common structure

I translated the strongest elements of the MVP into a repeatable learning arc:

• Establish why the skill matters.
• Introduce the core concept or mental model.
• Show what the skill looks like in practice.
• Give learners supported opportunities to try it.
• Provide feedback or a model for comparison.
• Increase the complexity or independence of the task.
• Connect the work to a credible career outcome.

The shared architecture gave learners a clear sense of direction and gave contributors a reliable place to begin.

I designed the model to be predictable enough that learners could focus on the skill rather than deciphering the course structure. Subject-matter experts and designers could then adapt the content, practice, and level of complexity to the needs of each subject.

The architecture created consistency in the experience without making individual courses feel interchangeable.

I defined the purpose of each stage and the relationship between explanation, modeling, practice, feedback, and application. UX designers and course developers helped translate that structure into the final interface and course experience.

Design response: Make every skill feel useful

A technical course can be accurate and still fail to feel useful.

The question I kept returning to was: What should learners be able to do after this lesson that they could not do before?

That question changed the role of activities within the course. Instead of adding practice after the content had been written, I designed lessons around:

• Reflection on how the skill could transfer to a current or future role
• Realistic workplace scenarios
• Decisions with credible tradeoffs
• Examples of how experts approach similar problems
• Guided practice followed by independent work
• Projects that produced tangible, career-relevant outputs

Career relevance became part of the course’s defining structure, not a message added at the end.

I designed activities and projects that asked learners to produce evidence of capability. The goal was not only to help learners understand a skill, but to leave them with proof that they could use it.

Product → System

A shared course architecture solved only part of the scaling problem. The team also needed a repeatable way to create, review, and improve the experiences inside each course.

Turn successful activities into reusable patterns

I converted recurring learning needs into adaptable design patterns.

I developed common approaches for concept explanations, worked examples, scenario-based decisions, guided practice, independent application, project instructions, and feedback. Each pattern had a specific purpose but could flex across subjects and levels of complexity.

Reusable patterns reduced the number of structural decisions the team had to make from scratch.

They also made reviews more focused. Instead of first determining how an unfamiliar lesson was supposed to work, reviewers could evaluate whether it accomplished its intended purpose.

Clarify the cross-functional workflow

I also clarified how work moved among instructional designers, UX designers, subject-matter experts, and reviewers:

• What each stage needed to accomplish
• What information contributors needed to provide
• Where review and approval should occur
• Which elements followed shared standards
• Which decisions required additional discussion or testing

The workflow helped the team distinguish content problems from experience-design and production problems.

I defined clearer stages, inputs, and review points so collaborators could identify what was actually causing friction. An inaccurate explanation, an unclear activity, and an inefficient handoff may look similar in a finished course, but they require very different solutions.

Use AI to accelerate the system

Generative AI created an opportunity to move faster. It also introduced a risk.

Without a clear architecture and quality standards, AI could simply help the team create inconsistency faster.

I used AI for bounded, repeatable tasks such as:

• Generating early variations
• Developing possible scenarios
• Reformatting information into established structures
• Identifying gaps and inconsistencies
• Supporting repetitive production work

Structured prompts and shared templates gave each output a clear purpose and made it easier to evaluate.

The course model made AI useful, not the other way around.

I designed the AI-supported workflow and defined where human judgment remained essential. Learning strategy, sequencing, audience context, factual accuracy, workplace credibility, accessibility, and final quality decisions stayed with the project team.

By pairing faster generation with clearer standards and deliberate review, we reduced repetitive work without lowering the bar for the finished experience.

Results

40%+ reduction in development

AI-supported workflows and reusable patterns made courses faster to produce without removing human review.

The system reduced repetitive drafting and one-off design decisions, allowing the team to spend more time on learner needs, subject-specific challenges, and final quality.

One MVP became a reusable model

The original course became the foundation for a broader portfolio.

The system combined a shared architecture, adaptable activity patterns, clearer quality standards, and a more consistent development process.

A clearer learner experience

Concepts, examples, practice, feedback, and workplace application became part of one connected learning arc.

Learners no longer had to infer how individual activities related to the skill they were building or the career outcome they were working toward.

Meaningful career outcomes

Across the broader Chegg Skills program:

• 76% of graduates reported higher earnings.
• Graduates who received promotions reported an average salary increase of $14,048.
• 66% said the program positively affected their careers.

The program connected learning to measurable career progress.

These outcomes reflect the broader Chegg Skills program, not the effect of any single design decision. They represent the purpose behind the work: helping learners turn new capabilities into meaningful career opportunities.

Reflection

The central challenge of this project wasn’t simply how to produce more courses. It was how to build a system that could grow without becoming generic.

Too little structure would have forced us to start over each time. Too much would strip away the context and specificity that makes a technical course worth taking. I found a compromise: standardize the foundation, vary the details built onto it.

The project also changed how I think about generative AI. Its value didn’t lie in the ability to produce more content, more outputs. Using AI was valuable because it effectively reduced repetitive, manual work once we already knew what “good” looked like.

If I continued work on this program, I would connect the system to more detailed behavioral data and user analytics: where learners paused or disengaged, which activity patterns produced the strongest performance, and how results varied across learner groups.