B2B • Product design

Designing an Agentic Quality Intelligence Platform

Motion-blurred streaks of yellow and teal with a small orange dot at the centre

Role: Founding Product Designer
Timeline: 3 months
Team: Arun Antony (Co-founder & COO)Shine Xavier (Co-founder & CEO)Sivakumar OK (Architect & Engineering head)Jacob Nelson (Engineer)

Overview

Making autonomous AI behavior understandable

PrimeCompass analyzes product behavior to uncover discrepancies between expected and actual system behavior. The challenge was turning complex AI-driven analysis into information teams could interpret and act on.

The challenge

Making system behavior observable and understandable.
Helping users investigate deviations from expected behavior.
Presenting complex analysis without overwhelming users.

The key insight

AI-generated findings become more useful when users can see what changed, understand why it matters, and trace the behavior behind the result.

Solution

Making autonomous behavior observable

I designed interfaces that expose discovered behavior, highlight deviations, and give users a clearer way to investigate how the product behaves against an expected baseline.

Discovery setup: the application URL and environment, personas and documents, with AI suggestions beside the Start discovery button

Turning behavioral change into a visual signal

I designed the Drift Map to show the relationship between a baseline and discovered behavior, helping users see where change was happening before investigating the details.

Discovery v5 overview compared against v11.4: module, feature and scenario counts, AI-generated insights, and the start of the Drift Orbit

System

Building a consistent language for the product

I established a reusable visual language for complex product states, aligning interface patterns and components to support consistency across the experience.

Brand primitives
Color, typography, spacing, radius, elevation, and foundational tokens.
Semantic aliases
Meaningful roles for surfaces, text, borders, feedback, status, and interaction.
Component tokens
Reusable components and interaction states mapped to the semantic system.

Final designs

Discovery in progress: AI suggestions next to a live discovery log and a summary of the modules, scenarios and nodes found so far

Modules and scenarios list beside the overview, with module, feature and scenario counts against the current baseline and AI-generated insights

Drift Orbit: modules plotted around baseline v11.4 by how far they have drifted, colour-coded as aligned, moderate or high drift

Outcome

From AI capability to a usable product model

My contribution went beyond individual screens. I helped shape how users understood, supervised, and acted on an autonomous system.

Observable AI
System findings presented through visual product states.
Traceable deviations
A clearer path from identifying drift to investigating its source.
Scalable design system
Established a design system that could support PrimeThoughts' broader product portfolio.

Reflection

AI needs to make its behavior observable

Designing AI products requires more than presenting generated results. Users need enough visibility into system behavior to understand discrepancies, question findings, and decide what to investigate next.