A research agent for a field designers can't easily look up

At our 56-customer clinic, every operator completed a 36-page survey by hand, with rating scales and open-ended answers, in 7 languages. That's about 2,000 pages to read, type up and translate before analysis could even start.

I worked with an engineer to use AI to digitize and translate the responses. To keep the data trustworthy we built in a human check: the AI rated its confidence in each reading, and I reviewed anything it flagged as low confidence. Facilitators also asked participants to write as clearly as possible, reducing errors from the start.

Once the clinic data was digitized, I saw that the same approach could solve a deeper problem.

Agricultural machines, devices and workflows are complex and unfamiliar to most designers. And our customers have a very small digital footprint, so designers can't simply look things up. Years of research lived in shared folders, useful only if you knew where to look.

So I'm building an agent that connects all our UX research files and results in one place. A designer can ask a question, like "What do operators say about armrest controls?", and get an answer grounded in real research.

1. 01

   ~2,000 handwritten survey pages in 7 languages

2. 02

   AI reads and translates each response

3. 03

   Low-confidence readings flagged for my review

4. 04

   Clean, structured data ready for analysis

5. 05

   A research agent that answers designers' questions

• Access: designers get answers from past research on their own, without waiting for a researcher
• Context: designers new to agriculture can understand customers they rarely meet
• Reuse: past studies keep informing new decisions instead of being forgotten in folders