Phase 1 MVP
The challenge was not just to create an AI agent. We needed to give users a clear way to create, train, configure, and validate an agent without losing track of where they were in the process.
For the MVP, we broke that experience into four connected parts.
X
PROBLEM
Users need a clear starting point to create and manage multiple AI voice agents without having to navigate through different parts of the product.
Y
SOLUTION
We introduced a central dashboard that brings existing agents, their status, and key usage information together, while making Create Agent the primary starting point.
Z
RESULT
Users can quickly understand the state of their agents and move directly into creating or managing an agent.
X
PROBLEM
Setting up a voice agent involves several decisions, including its purpose, industry, language, and voice. Without a clear structure, configuration can become difficult to follow.
Y
SOLUTION
We structured agent setup around these key decisions and kept them within a guided configuration flow, including support for multiple language and voice combinations.
Z
RESULT
The setup gives users a clearer path from defining an agent to configuring it for a specific use case.
X
PROBLEM
An AI agent needs relevant knowledge to respond effectively, but training data can quickly become difficult to manage as users add more files and content.
Y
SOLUTION
We created a dedicated knowledge and training workspace where users can upload, manage, and track the information used to train their agents.
Z
RESULT
Training becomes a defined part of the agent-building workflow rather than a separate or disconnected task.
X
PROBLEM
Users need to validate an agent's behaviour before exposing it to real conversations. Testing outside the product would make this process slower and harder to evaluate.
Y
SOLUTION
We designed an in-product playground where users can interact with the agent and test conversations while reviewing its language, preferences, and domain configuration.
Z
RESULT
Users can validate the agent in the same environment where they configure it, creating a tighter loop between setup → testing → refinement.
Prototype ideation
Before moving into detailed UI, we needed to validate the core flow and decide what information was essential at each step.
The initial sketches helped us explore the main tasks, feature hierarchy, and navigation without getting distracted by visual details.
What We Learned From the First Version
Observation of Phase 1
The first version was rolled out with stakeholders, team members and users through demo sessions. Over one sprint, we collected feedback to identify recurring issues in the experience and areas that needed refinement.
Feedback collected over one 15-day sprint
42%
of feedback identified opportunities for improvement
12%
of feedback was negative
33%
of observations mapped to recurring pain points
Sales Team Feedback After Testing
70%
Feedback captured
08%
of feedback was negative
32%
of user pain points recovered after feedback
Finding Patterns
What we heared
We grouped the feedbacks from the first version to separate isolated issues from recurring themes. This helped us identify which problems were related and where a broader product change was needed.
Emotional journey map
This map visualises a user's emotional journey while creating, training and testing AI voice agents on the Devnagri platform. Each stage represents their mental state, pain points, and emotional transitions throught the experience
Design System
What we heared
