Nikita.

Enter Password

Redesigned the AI chat experience to minimize user effort while making AI's reasoning more transparent

JAN 2025

my role

Product Designer — Designed the query feature, including the product thinking, UX and interaction design

Team

Cross Functional Team — Worked closely with the PM, 2 Developers, 1 QA

duration

3 months — time from research to design

Enter Password

impact

New design for the query was aimed at simplifying the user experience and helping users make informed decisions and choices with more flexibility

1/2

Reduced User Steps for querying questions

35%

Improved adoption and usage of the feature

context

We use RAG to generate answers from user-recorded information, but users must first select the right conversation due to the model’s limitations with large data scopes. Including too much data can dilute answer quality, making this manual step necessary but inefficient.

Currently, we use Retrieval-Augmented Generation (RAG) to generate answers based on a large volume of information recorded by the user. However, our current process requires users to first locate the specific conversations their question relates to. This adds friction, as users must manually identify the right context before they can ask a question, which can be time-consuming and limit the overall efficiency of the system.

opportunity

How might we simplify the AI query experience so users can easily find the answers they need with minimal effort, greater transparency, and an interface that feels intuitive and flexible?

Pain Points

Pain Point 1

Users need to know where to look.

RAG requires users to select the right source or conversation, but most users don’t know where specific information lives

Pain Point 2

Limited information can be retrieved

Only a small subset of content can be included at once; adding too much dilutes the response quality, forcing trade-offs in depth and coverage.

Solution

We designed a Query user flow with a semantic search layer to automatically surface the most relevant content while improving transparency around how the AI generates answers

We structured the design around five key elements to enhance clarity and efficiency. Kanban boards provide a quick visual snapshot of each group's progress, while progress data offers insights into overall team performance. The schedule helps track timelines and keep everyone on course. Recent activity highlights the latest outputs, enabling timely follow-up actions, and filters make it easy to sort and navigate content based on what’s most relevant.

understanding the technology

RAG has context size limitations, so we added a layer of semantic search to reduce user effort and surface the most relevant content for the question asked.

Key learnings

Design around familiar patterns (Jakob's Law)

Creating flows that align with user expectations and familiar mental models makes adoption smoother and reduces cognitive load.

Leverage technology to minimize friction

Using tools like semantic search can offload complexity from the user, streamlining the experience and allowing them to focus on their goals.

Transparency builds trust

Showing how and why the AI arrives at certain answers increases user confidence and engagement.