Vova Shlygin
Material Bank
A platform that allows architects and interior designers to order free material samples and get them delivered overnight.
As AI creative technologist in the product design department, I was focused on creating innovative user experiences by combining product design with prompt engineering, rapid prototyping and AI fine-tuning.
AI Inspiration
Role
Design Leadership, Prompt Engineering, Dev Management
Tech
CLIP, GPT, SAM, Milvus, Search API
After acquiring Designspiration and Architizer, Material Bank got access to the world's largest archive of architectural and interior design images. At the same time, the fact that the company stores more than 100,000 product sample SKUs created an unprecedented opportunity to build innovative inspiration experiences for architects and designers.
I was leading this project by designing, prototyping and building sandbox features together with an ML engineer, a front-end engineer, and a product designer. We were also supported by a product researcher from the primary Material Bank product team.
Play with it here. Login "as". Password "is"
Ideas Search
This feature enabled interior designers to create image collections with a single text request. The LLM-based solution I've designed comes up with ideas as images instead of returning direct query matches. The system delivers instant moodboards. The ideas are also used as filters that allow users to quickly find inspiration they are interested in. We've built the entire thing as a tech layer on top of the primary product search engine.
I've designed a prompt that is able to differentiate between space descriptions ("pink hotel lobby") and categories ("modern chairs"), and output a uniform JSON with ideas and search query descriptions. We also auto-assigned icons to specific keywords for easier navigation.
One-Click Sampling
Generative Auto-complete
This feature enables users to build elaborate search requests with generative phrase autocompletion. Our solution builds suggestion options without any historical search data.
AI Paint Visualizer
Role
Prompt Engineering, Research
Tech
Stable Diffusion, Multi-ControlNet, SAM
→
To implement "try before you buy" flow for a B2C product we designed a tech stack for changing materials in any image. We built proof-of-concept prototype for paint products, and designed a solution for tackling complex objects such as plants.
→
→
AI Writer
Role
Design Layouts, Prompt Engineering
Tech
GPT
AI writer is an internal copyrighting assistant based on corporate guidelines created by Material Bank copy lead. This tool enables marketing, design, product, and internal comms teams to generate their own content.
The solution is based on a variable GPT prompt. Every UI input is assigned to a specific part of the prompt, so by interacting with the front-end users change the prompt slightly.
Homepage Personalization
Role
Design Layouts, Tech Research
Tech
Search API
I developed a personalized user experience for Material Bank, performed tests with user data, and crafted a custom homepage experience. The fundamental concept behind Material Bank's customization is that each user's activity influences the content displayed on the homepage.
Material Bank operates a business model that profits from providing user data and leads to brands, so we needed to encourage users to share more data. For that reason we decided to create a project completion block that affects the entire feed when our user fills in details about their project.
Samplize Redesign
Role
Design Layouts, Project Management, Development Management
Samplize is online marketplace for sampling paint overnight. I revamped the entire online user journey, hired a team of developers, and oversaw the creation process. The relaunch led to revenue increase in the fourth quarter of 2022.
Thank you!
Questions? Don't hesitate to reach out.