K1st BUILD
Building AI models from human expertise
Company
Aitomatic
Position
Founding Product Designer
Year
2022 – 2023
Market
APAC & US, B2B
Problem
Manufacturing teams lost time and momentum building AI models, stuck with incomplete datasets and heavy prep work.
Solution
Built K1st BUILD, a knowledge-first app that captures expert input to accelerate model development without heavy data reliance.
Impact
•
Cut data prep time by 30%
•
Boosted model accuracy by 15%
•
Improved alignment between engineers and experts by 20%
Problem
Manufacturing companies depend on accurate AI models for tasks like predictive maintenance and operational optimization.
AI engineers were spending significant time cleaning and prepping data, only to achieve modest results due to gaps in datasets. At the same time, industrial domain experts, with 30 to 40 years of invaluable operational insights, struggled to effectively pass on their knowledge. Their wisdom remained across documents, informal conversations, and personal notes.
At Aitomatic, the opportunity was clear: how can we build accurate AI models when the data itself is imperfect? My role as Founding Product Designer was to create a radically different solution: AI model-builder that started from human expertise instead of raw data alone.
Design principles
With the insight from leadership team, I framed the design around the knowledge-first approach. Our core belief was that structured human knowledge could bridge data gaps, boosting model accuracy and shorten development cycles.
I envisioned a system that allowed engineers and domain experts to capture and reuse knowledge easily, creating a powerful loop of continuous improvement. By elevating human expertise within the AI workflow, we could achieve accuracy, speed, and team alignment.
Design process
I began by collaborating with CNA’s AI engineers and domain experts. Through in-depth interviews, and workflow analysis, I mapped how experts communicated insights and how engineers add these insights into models. I discovered critical friction points where knowledge was lost or misinterpreted, pointing out where design could deliver immediate value.
Using research results, I created fast prototypes to align with PMs and Sales, and later the users to refine the platform's usability. Each iteration streamlined the process, reducing friction between domain experts’ knowledge-sharing and AI engineers’ model-building tasks.
What we built
K1st BUILD, was designed around two user personas: domain experts and AI engineers.
For the domain experts, the platform offered a simple yet powerful way to share their expertise. They could document knowledge with chats or uploads, refine schemas, and validate model outputs. Their operational insights became structured, reusable knowledge that filled gaps left by imperfect raw data.
For AI engineers, I crafted a streamlined workflow that dramatically reduced data preparation time, enabling them to use structured insights to build models. Clear data schemas, intuitive model-building interfaces, and easy iteration paths made tedious tasks now efficient and precise.
The design prioritized effortless collab between the two, facilitating a feedback loop. Engineers could refine models using inputs from domain experts, who in turn could see their knowledge reflected in model improvements.
Impact
K1st BUILD reshaped how manufacturing teams approached AI model development.
By blending human insights into model creation, data preparation time dropped by at least 30%, freeing AI engineers to focus on high-impact tasks rather than data wrangling. Across critical use cases like predictive maintenance, model accuracy improved by up to 15%, accelerating project timelines from months down to weeks.
Beyond technical gains, the most profound impact was 20% improvement in team alignment, attributing this to the collaborative workflows inK1st BUILD. Engineers and domain experts had a shared language built around new ways of building.
