Product Design
Design Strategy
Product Designer
San Francisco, CA
2023.05 -Current
01
Hive’s AutoML platform allows customers easily train, evaluate, and deploy customized machine learning models. Beginning with custom data, AutoML platform guides the users through the process of creating a fully-functional model accessible via an API endpoint just like other Hive pre-trained models.
Datasets are the very first step to building an AutoML model. A dataset represents the information that you would like to use for model training or to create an embedding. On our Datasets page, you can preview and edit your information before you use it to train or augment a model.
An AutoML model is a custom machine learning algorithm trained using your . Models are created using Dataset . After a model has completed training, you can deploy that model to instantly begin using it by submitting API requests as you would all other Hive models.
An embedding is a vectorized form of a dataset which can be used to lookup text data. Embeddings can be queried directly to search for text that is similar to the provided input or used to augment LLM.
02
- Rising demand for custom AI models.
- Aligns with Hive's goals to expand AI services.
- Streamlines internal model training processes.
Recognizing the expanding AI industry, we observed an influx of new customers and users gaining a deeper understanding of AI. As they expressed the need for customized AI models to suit their specific use cases, we noticed a limited number of competitors in this niche.
Being a major player in the pre-trained AI model market, we seized the opportunity to not only expand our services but also enhance the experiences of our existing customers. Internally, this could also serve as a tool to expedite our team's model training process.
Given the novelty of this industry, even our internal team requires time for exploration and education on the technology. This process is essential to truly discern our capabilities and formulate a well-informed approach.
As this is a highly innovative product and the entire industry is still in the exploration phase for this specific niche market, it's crucial for us to pinpoint the most viable target audience and understand their needs.
Timing is a significant factor in the competitive landscape—finding a way to efficiently develop a product and conduct testing within tight time constraints is key.
Identifying an appropriate marketing channel and effectively onboarding customers to the product stands out as a significant challenge in the design process.
03
After analyzing potential competitors in the market, it's evident that none of them specifically target the moderation use case or hold a distinct advantage in this area. Furthermore, while our tier 1 competitors dominate global market share, tier 2 competitors lack a significant emphasis on AutoML products. This positions Hive to potentially become competitive among tier 2 competitors.
As pioneers in the AutoML market, we aspire to pinpoint early adopters as our initial target customers for our early-stage product. Leveraging their influence, we aim to broaden our market reach and gradually refine our product for the wider public.
Thankfully, Hive has cultivated a reputation in Generative and Moderation AI Models, attracting customers who express their interest or specific needs for customized solutions. Capitalizing on this advantage, we've identified them as our early adopters for the product and strategically utilized our existing platform to market it.
Hive’s AutoML platform allows customers easily train, evaluate, and deploy customized machine learning models. Beginning with custom data, AutoML platform guides the users through the process of creating a fully-functional model accessible via an API endpoint just like other Hive pre-trained models.
We looked at Hive's brand and how well the AutoML Dashboard fits our mission. After that, we established our post-redesign goals. Our focus is on creating a user-friendly interface with versatile features to enhance the overall user experience in key dashboard areas.
Our vision for this product is to serve as an all-in-one model training platform, offering users a seamless and comprehensive experience from start to finish in their model training journey.
We aim to create a comprehensive dashboard, allowing customers to effortlessly customize their models with low knowledge requirements.
Our objective is to optimize certain features specifically for our primary target audience, focusing on enhancing the experience for content moderation and generative AI customers.
We aim to empower users with more control over the training process, allowing them to tailor it according to their specific needs and preferences.
04
We opted to initiate our project by crafting a Minimum Viable Product (MVP) with the a single simplest flow, essentially using it as a foundational framework for further development. This approach allows us to efficiently test the initial user experience and assess our technical capabilities.
To gather more firsthand insights into user behaviors, we conducted qualitative research, concentrating on a select group of interviewees with backgrounds closely aligned with our target user demographic.
After conducting the interviews, we constructed a comprehensive user journey map that encapsulates all pain points, expectations, and emotional responses throughout the dashboard's various stages. This map serves as a valuable tool in understanding user experiences, enabling us to pinpoint critical areas for improvement and optimization.
05
Incorporating insights from user research, we can expand the product by introducing additional flexible steps into the core flow, enhancing its features and scalability.
Building upon the core flow as a foundation, we broke down tools and content into four crucial sections: Datasets, Models, Embeddings, and supplementary dashboard features. This strategic framework empowers us to extend the dashboard while preserving the product's essence, minimizing the risk of user disorientation during incremental feature updates.
In our quest to optimize user experience, we've organized information into specific pages. Additionally, we've strategically place similar information in one place, reducing cognitive load and minimizing screen movements. This thoughtful arrangement enhances overall productivity.
Moreover, we've improved the page hierarchy and contrast while limiting each page to a single key action button. This approach ensures users can readily identify the most crucial information they need.
The moderation dashboard previously featured its own color and component system, which had become cluttered over time. With the ongoing design updates, our objective is to systematically restore the brand aesthetics and structure of the dashboard.
Acknowledging the diverse user groups utilizing various screen sizes and devices for task management, we leverage the card component, allowing the components to fluidly adjust within the guidelines.
06
the creation of a dedicated section empowering users to effectively manage their datasets for model training and embedding. Enabling users to effortlessly access, review, and edit their datasets is integral to enhancing the overall user experience. Now, users can seamlessly prepare, manage, and utilize their data, all within a unified and streamlined platform.
The model section is intricately crafted to cater to customers, offering them the tools to train, evaluate, and deploy their custom models seamlessly. Additionally, we've streamlined the evaluation sections, providing customers with a quick and efficient assessment of their trained models. The introduction of the new playground feature, customized for various model use cases such as content moderation, now allows users to actively test their custom-trained models.
The embedding section, commonly referred to as RAG (Retrieval Augmented Generation), allows the customers to create a vectorized form of a dataset which can be deployed to allow quick lookups of similar data such as images or text. This functionality significantly enhances the accuracy of large language model outputs.
07
We've revamped the moderation dashboard design, ensuring seamless integration with the AutoML dashboard. This not only enhances overall cohesiveness but also functions as a compelling advertisement for our moderation customers, encouraging them to consider upgrading their models for an even more advanced and streamlined experience.
Given the complexity of our product and the fact that many users are new to AI model training and machine learning, we've meticulously organized our documentation. This ensures that users can easily navigate and find all the information they need from the beginning to the end. We've taken care to explain terminology and concepts comprehensively within the documents.
In refining our marketing approach, we initiated the practice of sending marketing emails and updating our blog ahead of each release. Simultaneously, we launched a website to furnish prospective AutoML customers with vital information in advance of their engagement.
08
Initiating a 0-1 product demands a distinct approach from redesign projects. Essential to this process is identifying core needs and adopting effective customer engagement strategies. Common sense is a valuable tool for making initial assumptions and laying a solid foundation. Moreover, solving for a general audience not only addresses broader needs but also provides valuable insights and benefits for professionals in the field. This dual focus enhances the overall impact and success of the project.
A comprehensive overview of my pivotal contributions at Hive, encompassing brand design strategies, product design, and marketing designs.
Kodak Vision is a rebrand project that focused on revitalizing the brand, Kodak, and identifying new product opportunities. It also entailed designing a VR creative workspace.
DiDi Switch is a new service model designed for one of the major transportation service company—— DiDi Chuxing.
