
How might we design an art foundation that excites and delights while amplifying the voices of marginalized communities? How might we use a challenging sloped site in Los Osos California to entice movement and interaction throughout the project? How might we create a structure that is raw, digestible, and understood by its sense of place?
yolk is a public art foundation that amplifies the voices of black artists by displaying the works of El Enatsui and Senga Nengudi. The project is derived from simple geometries that have been translated by their implementation within the site. The raw, digestible, and naked approach creates an intimate atmosphere where visitors feel compelled to interact with the art along their journey through cavernous spaces. Spaces are influenced by and capitalize on the use of light, illuminating routes on navigation and showcasing the art in it's finest form.
The site came with it's own unique qualities being placed far from the urban city of San Luis Obispo and presenting topographic challenges with a slope above 10% with an abundant coverage of native trees and foliage that we sought to protect. Our team thought about who would be likely to visit this site and who we could be designing for. Additionally, this project was designed amidst the global pandemic of COVID-19 as well as what some might describe as the forefront of Black Lives Matter Movement. These conditions no doubt impacted our own personal goals of the project.
%203.jpg)
Just as the work of our chosen artists embraced an analog attitude towards their work, we did the same. From this point our project became dedicated to the reuse of materials and using analog methods of representation, just like our artists did. The following is a logo we designed for our project. The charcoal drawn symbol contains two layers: the shell representing the outside protective structure of our building, and the yolk, which is the lively heart of our building: the art.
%201.jpg)
The next phase of the design process was to bring our formal definitions and goals for yolk into something tangible. Using analog strategies, we sketched, painted, and crafted informal explorations around our goal for the art foundation. How can we hold something that is raw, naked, and digestible? How can we use light to illuminate and drew attention to the spaces created?

Some of the themes myself and Emma pulled from our early discoveries were the layers that come from a hard exterior shell and the gooey interior yolk of an egg. Taking that object a step further, we brought it to the site through plaster and rockite explorations. Emma's discoveries on the left used an egg to guide an additive approach on the site, while my explorations on the right were integrated and cut into the site: a void or subtractive strategy. Emma and I's analog form experimentations were ultimately combined into a hybrid strategy to create the overall form of the project. Integration to and sculpting of the site were extremely important in our narrative.

Using different approaches, Emma and I ultimately arrived at a hybrid composition for the project: both additive and subtractive forms cutting into, sitting on top of, and speaking to the site. Both the forms themselves as well as the strategy by which we sculpted and built earth on the site encourage circulation through the project and help to lead users on a specific journey in order to create an interactive and intimate experience.
Like any successful design project, 2 dimensional drawings are used to better understand parts within our project at the human scale. Here we unpack how to think about moving inside of the building, defining program, and better understanding how to use light to encourage movement throughout the spaces.

In crafting our floor plans, Emma and I thought about which spaces would be best suited for particular programs or pieces of art. For example, on the long curved walls, El Anatusi's textile quilts were best suited for those spaces as Nenguidi's tensile installations were best suited in our double height gallery space as well as in our structural pit located east of the project. The project contains multiple gallery spaces, a cafe, and educational space on black art and cultural history in the San Luis Obispo area and beyond. Users wander and curve through spaces being led from one part to another by the form, light, and the artwork.

Designing through use of section was one strategy that Emma and I used to understand the differing levels and thresholds between spaces. Through creating sections we better understood how to best use daylight throughout the building and which areas would be darker or lighter. These sections also show our attempt in understanding how the project ties into the ground and moves users from one level to another. We understand the spaces at a human scale and how to create a sense of awe and by manipulating the ground plane and height of each individual space.


Towards the end of our project began to attempt to understand our project on structural level. We explored various construction methods and settled on a layered steel grid construction assembly. It helped us to better understand how this system works by breaking it into layers and looking at in section too.
Leading the Mini Cart feature at Autodesk meant confronting an embarrassing reality: we didn't have a modern shopping cart experience. Customers couldn't configure products, adjust terms, add multiple items, or even access their cart from the universal header. The entire flow was "select product → go directly to checkout"—no flexibility, no discovery, no optimization opportunities. This wasn't just a UX problem; it was a massive business limitation. When customers did manage to buy multiple products, their average order value was 4x higher, but only 1.5% of orders contained multiple items. We had no recommendation system, no way to encourage cross-product exploration, and checkout conversion rates that reflected the friction. I was tasked with building a complete shopping experience from scratch, including the machine learning infrastructure to power intelligent recommendations.
I led a cross-functional initiative to transform our basic e-commerce flow into an intelligent shopping platform worthy of modern customer expectations. The "Mini Cart" name reflected our vision: bring key checkout capabilities into a smaller flyout panel where customers could finally configure their purchases like they expected. Phase I focused purely on the foundational shopping experience—cart persistence, product configuration, and multi-item management. Phase II introduced the machine learning magic: a collaborative filtering model that I helped design alongside data science and engineering teams, complete with business rules layers for regional preferences and marketing campaigns. The vision extended beyond algorithms to incorporate user entitlements, behavioral signals, and purchase history for truly personalized experiences. Through careful measurement and iteration, we proved that sophisticated recommendation systems could work in B2B environments, qualifying traffic more effectively and driving meaningful cross-product adoption.
As the lead product manager for the Mini Cart feature, I identified a fundamental gap in Autodesk's e-commerce experience. Believe it or not, we didn't have a modern shopping cart experience—customers couldn't configure products, adjust terms, add multiple items, or even access their cart from the universal header. The name "Mini Cart" came from our vision to bring key checkout capabilities into a smaller flyout panel, creating the modern shopping experience our customers expected.
The data revealed the scope of missed opportunities. While our baseline checkout conversion sat at 18%, we were failing to capitalize on customers' natural shopping behaviors. The most telling insight was around multi-product purchases: when customers did buy multiple products, their average order value skyrocketed from $697 to over $2,700—nearly 4x higher. Yet only 1% of orders contained multiple products, suggesting massive untapped potential.


The User Experience Gap: Customers were being pushed immediately to checkout after adding items to their cart, missing opportunities to explore complementary products or adjust their configurations. This created a linear, inflexible experience that didn't match modern e-commerce expectations.
The Technical Challenge: We had no recommendation system whatsoever—customers went directly from product selection to checkout with zero opportunity for discovery or cart optimization. We needed to build a machine learning-powered recommendation engine from scratch that could understand user patterns across our complex product portfolio. Perhaps most alarming, 54% of our APMs expressed only moderate confidence in their ability to transfer their current skills to other domains or teams. This suggests they were developing highly specialized, context-specific capabilities rather than the versatile, transferable product management competencies needed for career growth.
The vision became clear: create an intelligent, dynamic shopping experience that would serve both customer needs and business objectives. Rather than treating the cart as a simple container, we would transform it into an intelligent advisor.
The Machine Learning Foundation: I partnered with our data science team to implement a collaborative filtering model that could identify patterns between user behaviors and product relationships. This required close coordination across content, data science, and engineering teams to build something completely new for our platform.The model incorporated business rules layers that I helped define to account for regional preferences and specific marketing campaigns that pure algorithmic recommendations might miss. This hybrid approach ensured our recommendations were both data-driven and business-intelligent.

Vision for the Future: The long-term roadmap included incorporating user entitlements and behavioral signals—if a customer had a trial, their purchase history, or had visited the Revit page dozens of times, the system would learn and adapt. This contextual intelligence would create truly personalized experiences, though we were building toward this capability incrementally.
Strategic Objectives:
Minimize friction while maximizing personalization opportunitiesExpedite business value creation through increased AOV and multi-product adoptionEmpower customers with intelligent discovery and configuration tools
Phase I: Foundation BuildingThe first iteration solved the core shopping experience gap. We built the Mini Cart flyout that provided persistent cart access across product pages, promotions, and navigation. For the first time, customers could configure products, adjust licensing terms, add multiple items, and continue shopping—all without losing their progress or being forced immediately to checkout.This phase was entirely focused on creating the modern cart experience that was previously non-existent. No recommendations yet—just the fundamental shopping capabilities that customers expected.

Phase II: Intelligence IntegrationThis is where the machine learning model came to life. The mini cart became a platform for real-time, contextual recommendations powered by our collaborative filtering algorithm. Unlike static suggestion engines, our system could adapt to each user's unique context:
Dynamic Product Recommendations: The ML model analyzed current cart contents and user patterns to suggest complementary productsCross-sell Opportunities: Intelligent identification of product combinations that delivered higher valueContextual Offers: Store-wide promotions and collection upsells targeted to customer interest patterns



The recommendation visibility varied by region due to privacy preferences—47% of US customers saw recommendations versus 31% in Japan and 20-30% in EMEA. However, engagement was strong: 1.9% of customers added recommended products to their cart, and those who did had a 13% conversion rate.

Immediate Results: Within 180 days of Phase I launch, we processed 67K orders with an improved AOV of $697 (up from $684). More importantly, multi-product cart adoption increased from 1.5% to 1.6%—a seemingly small change that represented significant revenue impact given the 4x AOV multiplier.
The Qualification Effect: The most significant discovery was how Mini Cart improved traffic quality throughout the funnel. While total conversion didn't dramatically increase, customers who engaged with Mini Cart were far more qualified prospects.


Continuous Optimization: Based on initial learnings about add-then-remove behavior, we conducted A/B tests to reduce customer confusion. By defaulting the recommendation section to closed, we improved measurement accuracy while maintaining engagement quality.
Looking Forward: The foundation enabled expansion into collection upsells and broader recommendation strategies, with Phase III planned to support Flex products and align with our Order-to-Pay checkout modernization.


Conversion Quality Improvement: The checkout conversion rate jumped from 18% to 38%—more than doubling—for customers who used Mini Cart features. This suggested we weren't just adding features; we were fundamentally improving the shopping experience for engaged customers.
Behavioral Insights: Nearly half of users actively engaged with Mini Cart elements, with 30% clicking through to checkout. The data revealed distinct engagement patterns:Customers who used quantity adjustment features had significantly higher AOVsThose who switched users or continued shopping showed even stronger purchase intentEven customers who initially closed the Mini Cart had notable eventual conversion rates
Cross-Sell Success: Multi-product orders generated over $1M in incremental revenue, with average order values nearly 4x higher than single-product purchases. The most popular combinations validated our ML model's effectiveness: AutoCAD + Revit, AutoCAD + AutoCAD LT, and Revit + BIM 360 Pro leading cross-product adoption.