Solution
Provides Code Support as your AI pair programmer
Helps sending Automated Emails and setting up Meetings
Manages your Helpdesk tickets
Instantly get your Meeting Summaries
Helps track your to-dos by sending reminders
NVAssist is a workplace support chatbot powered by Generative AI, which answers users’ queries and streamlines daily tasks as a personal assistant. It is a centralized platform that integrates with various enterprise workplace tools, creating a comprehensive ecosystem and allowing users to access workplace tools within a single interface.
Timeline : April 2023 - Dec 2023
Team : 4 Members
Role : Product Designer
Contributions: Defining User Flows, UX Design, Wireframing, Prototyping, Visual Design, Research
Tools : Figma, Adobe Creative Suite, Miro, Qualtrics, Google Suite
26% find it challenging to switch between multiple apps
21% need strategic tools for managing tasks
25% feel time is wasted in accessing required data
.png)
Our research aimed to explore to understand the below
What are users' expectations and current perceptions of chatbot as a solution at workplace?
What design aspects of AI chatbots would be most likely to build and maintain users' trust over time?
What accessibility considerations should be included in the design of AI chatbots?
We conducted 12 interviews and performed contextual inquiry to understand what they think about LLM chatbots.
1️⃣ Step
Semi-structured formative interviews
To understand users’ expectations and perception of LLM chabot, traditional customer support chatbot and factors contributing to their trust in these chatbots.
2️⃣ Step II
Contextual Inquiry
We wanted to know if verbal feedback reconciles with behaviors of their interactions with
chatbots (users say == what users do)
Total Participants - 12
1. Students - 3
2. Other Industry professionals - 4
3. Nvidia professionals - 5
.png)
3️⃣ Step III
Quantitative Survey
We still wanted to probe upon the design decisions before we start prototyping. So we decided to launch a survey to further justify the findings and and better inform the design decisions.
The survey questions were based off of the initial interview findings. We included 3 types of
questions:
1. Overall preference question to let user indicate their favored design options.
2. We included Likert scale to capture possible nuances beyond mere liking or disliking,
3. We also added an open-ended comment section for users to provide additional qualitative feedback.
In the end, we received 42 responses and performed statistical analysis to the data, which added more confidence to the interview findings, and which takes us to our final research findings
We came up with 3 themes based on user interviews -
Key Finding 1: Errors or hallucinations discourage user trust
Key Finding 2: Users can’t determine how AI chatbots arrive at the responses
Key Finding 3: Lack of Human-likeness with chatbots
Key Finding 4: Lack of contextual understanding and flexibility with chatbots
Key Finding 5: Users not only look to AI chatbots for guidance but also expect them to actively implement the suggestions and instructions they provide
Turning research insights into design opportunities!
From our research we discovered that users are expecting the chatbot to provide Targeting Responses such as - having appropriate tone/style, encouraging users to complete a task or making a decision were solved via Prompt Engineering.
Other expectations such as having more trustworthy experience by providing visual cues, design elements were solved with UX Design Features.
Complete Solution 🎉 = Prompt Engineering + UX Design Features
.png)
With all our findings in place our aim was to ensure that the prompts align with our key discoveries and contribute to issue resolution.
We continuously refined the prompt design, eliminating redundant instructions while maintaining the consistency of our key findings. We tested our prompts with Nvidia's open-source HR support information pages.
We were able to address -
1. Providing Source Links with the chatbot's responses
2. Managing the Verbosity of the chatbot's response
3. Providing Chain of Thoughts so that user knows how the response is being formed
4. Having Empathetic Tone to the responses
5. Asking follow-up questions if user doesn't provide sufficient data
Below is an example of prompt design that we did-
1. Feedback Mechanism
2. Displaying Source
Links
3. Having Multimedia along with textual
responses
1. Onboarding tutorial
2. Speech-to-Text input/ Text-to-Speech conversion
3. Smart Suggestions
4. Chat History
Assistance with:
1. Providing Coding
Support
2. Setting up Meetings &
Generating Automated Notes
3. Drafting Emails Draft
Raising Helpdesk Tickets
1️⃣ Low-Fidelity Designing
2️⃣ Usability Testing
3️⃣ Iterations based on key findings
We conducted Usability Testing with 15 participants to understand what are their initial thoughts about the chatbot what do they like/dislike with that we made improvements to our designs.
Expectation : Users expected the AI chatbot to help with their personalized productivity
Design Implication : Integrating personal meeting calendar, to-do list & notification
Using smart task suggestions & quick confirmation
Expectation : Users preferred real-time feedback between their chat and the bot’s task draft along with integrated enterprise UI solutions that helped them trust the chatbot more.
Design Implication : Enabling direct chat with auto-fill, designing split screens for instant chat and task feedback, using company’s workplace tools integrated directly with the chatbot.
.png)
Click to view full Figma work file here ↓
Together Everyone Achieves More (TEAM) 😊
A comprehensive project relies on the contributions of each team member we 4 of us with diverse talents worked together for making this project a success after learning from failures. We continuously improved within a team, driving future project success.
Passion Fuels Innovation ✨
We all loved work and the project that we were working on. It helped us push boundaries, and explore new ideas.
2 Research Papers coming on the way! Stay Tuned...! 💡
.png)