I make AI legible to humans. And sometimes the other way around.
3.5 years designing enterprise AI at Fractal Analytics. Side projects at 1am. National hackathon judge. Bird photographer.
Applying to Bristol · Birmingham · City · Glasgow
Edinburgh · York · Newcastle
Unified 10+ Cogentiq enterprise AI products under one design language system — 173 tokens, 42 components, enabling faster feature velocity, consistent UX, and shared governance across 6 designers and 10+ products.
Context & Challenge
Fractal Analytics' Cogentiq platform is a suite of 10+ interconnected enterprise AI products serving hundreds of organisations. By mid-2025, the design ecosystem was fragmented — each product team had evolved its own patterns. Buttons looked slightly different. Colour systems didn't align. Typography varied. Developers received ambiguous handoffs. Designers spent cycles recreating components instead of solving user problems.
Fragmentation had real impact: slower feature velocity, inconsistent user experience, duplicated effort, and design debt accumulating faster than it could be addressed.
Leadership assigned us to build a unified design language system that could scale with the organisation and guide hundreds of designers and developers.
Research & Discovery
Rather than imposing a system top-down, we started by understanding the problem from the inside — stakeholder interviews with product designers, developers, and engineering leaders across Fractal's Cogentiq division.
Key Findings
Designers actually wanted constraints. Constraints forced creativity on problems that mattered. Freedom on trivial choices felt like cognitive overhead. This shaped our entire approach.
Design Process
Starting with an Audit
We audited patterns across all 10+ products — every button, input, card, and modal variation. We found dozens of 'almost identical but slightly different' components that had evolved in isolation. This audit grounded us in reality and became our foundation.
Token-First Architecture
Start with tokens, not components. Tokens are the grammar of a design system — get the structure right and components follow naturally.
Iterations
Governance Layer
Teams could request new components; the design system team reviewed them. If truly new (not a variant), it shipped. This prevented chaos while staying flexible.
Solution & Impact
What Shipped
Measured Impact
Designers could focus on real problems. Developers knew exactly which token to use. Design critiques became about strategy, not debate.
Reflection & Learning
This project taught me that design systems aren't about design. They're about governance, change management, and how organisations work.
What I'd Do Differently
Building a design system requires research rigour beyond typical design. You're designing for designers and developers as users — different research, different metrics, different success measures. Systems design is underexplored territory.
Designed UX for an AI-powered underwriting platform serving a major insurance client. NDA applies to outputs — this documents process and design decisions only.
UX design for a machine learning predictive maintenance platform. Content being compiled — check back soon.
Design systems · Error states · High-stakes interfaces
IndiaHCI · AI accessibility · Community design
StayPut origin · Attention design
B.Tech CST · UEM Kolkata · CGPA 9.45
IxDF — Information Visualization
IxDF — AI for Designers