Pienso NLQ
Text-supported answers to users’ questions about their data
- UI
- UX
- AI/ML
- CUI
- 0->1
NLQ is a new feature in Pienso Explore that lets users ask questions about their documents.
NLQ is a conversational interface for asking questions about your data and getting reliable, document-supported answers.
Team
Design Will Crum
Development Brian Cort, Mathew Maradin, Felipe Balduino Cassar
Background
Post-ChatGPT, UX expectations were evolving quickly. And Pienso’s Explore MVP already had the vector architecture to run “RAG”-powered LLM responses.
Objective
Let users ask natural language questions about their data and deliver accurate, document-anchored responses.
Process
Kicked off Jan 2024, MVP established by May 2024. First I confirmed the MVP question set and established Pienso’s relevant design principles. Then it was: wireframe, iterate, build, test, and iterate some more.
Challenges
- No false answers! – Leadership was determined to limit hallucinations, to limit liability and build user trust.
- Finite context window – Besides the query, the LLM receives part/all of the relevant docs. How to constrain?
- AI as a tool – NLQ should empower human expertise, not replace it.
Background
- Query templates – Mad lib-style fill-in-the-blanks to constrain MVP users
- One-shot responses – No chatting back and forth, just a single Q&A
- Cited sources, centralized docs – NLQ’s answers quote, cite, and highlight the source docs, which remain central in Explore’s layout.