Prop AI

Experimenting with LLMs and RAG technology to facilitate proposal creation by internal teams at NI

snapshot

Summer 2023: 3 months
Company: NI
Users: Internal Proposal Engineers focused on RFQs for industrial test systems
Team: Long Term Innovation squad focused on internal applications of AI and ML for internal use cases, comprised of 3 engineers and 2 designers
Role: Product Designer
Outcome: A gpt-3.5 and 4 based prototype that enabled Proposal Engineers to more easily summarize, translate, and query RFQ documents.
Latest Concepts:

PROCESS

The LTI team (2 designers, 2 full-stack engineers, 1 data scientist) met weekly to listen, collaborate, and get feedback from a few representative end-users, engineers from NI’s Proposal team. These Proposal Engineers evaluate requests for quotes (RFQs) from customers for complex test systems (like a car manufacturer who needs a test system for their battery manufacturing facility), and create proposals for solutions that NI can provide.

Our first few weekly sessions were focused on learning about the existing workflow, tools, inputs and outputs these specialized engineers use for this complex task.

After that, we…

LEARNINGS & OUTCOME

The Proposal Engineers found that using LLMs to query RFQs an interesting yet imperfect solution, with one saying: I need facts not philosophy!

They appreciated the layout and intent for the UI to eventually help users find the source of the LLM-provided answers in the source RFQ.

Process-wise, this tight collaboration between design and development was appreciated by all, and the LTI team began collaboration on additional opportunities hypothesized as good candidates for LLMs, like ADAS sensor emulation.

Unfortunately, our time was cut short in the fall of 2023 with a round of layoffs attributed to the recent acquisition of NI by Emerson. Both myself and the other designer on this team were let go. I enjoyed working with emerging technologies in close collaboration with builders and users and will continue to advocate for rapid prototyping where appropriate.