Building design infrastructure
where none existed

The Situation

Designers used one set of color names, engineers used hex codes, and PMs referenced brand guidelines that didn't match either. Components were being rebuilt from scratch for every feature because there was no token system, no source of truth. The four merging airlines made this exponentially worse — Vistara's “primary blue” was not Air India's “primary blue.”

My Approach

I audited every screen across all four airline apps — 200+ screens — catalogued every unique value (colors, spacing, type sizes, radii) and mapped them against the new unified brand guidelines. Instead of proposing a token system top-down, I built a working prototype first: a small set of semantic tokens applied to one real feature, then demoed it to engineering leads to show the reduction in back-and-forth.

The Decision & Why

I chose semantic tokens over primitive-only because the merger meant brand values would keep changing — semantic naming let us swap foundations without touching component code. This was a harder sell initially because it required engineers to learn a new naming convention, but I built the case by showing how many production bugs traced back to hardcoded values. Three engineering leads signed off after seeing the prototype reduce spec-to-code discrepancies in a pilot feature.

The Situation

Design QA was happening through screenshots in Slack threads. Designers would eyeball implementations and flag issues in comments. With 10M+ annual users, pixel-level inconsistencies at scale meant degraded trust in a brand trying to reposition itself as premium. The team was spending ~30% of sprint time on QA back-and-forth.

My Approach

I scoped the most common QA failures — spacing mismatches, wrong color tokens, font-weight errors, missing states — and realized 70% of issues were mechanically detectable. Instead of advocating for a better process (more meetings, more checklists), I built a tool. Pixel Radar is a Figma plugin I designed and coded (5,000+ lines) that automates visual QA against the token system.

The Decision & Why

I built it as a Figma plugin rather than a standalone tool because adoption was the real challenge — the team was already in Figma. I chose to scope v1 tightly: only token compliance checks, no subjective quality metrics. This let me ship in 48 hours during an internal hackathon (which we won, in a Microsoft-partnered event), prove value immediately, and iterate based on real usage. The plugin now serves 450+ daily active users across the design org.

The Situation

Leadership wanted “AI in the product” but there were no internal frameworks for when AI added value vs. when it added complexity. The search experience — a core user flow touching millions of queries — was identified as the first AI integration point. But the risk was high: a bad AI search experience on a travel booking platform directly impacts revenue.

My Approach

I ran a competitive analysis across 12 airlines and adjacent travel platforms to map the landscape of AI-assisted search. Then I framed the problem not as “add AI to search” but as “reduce time-to-relevant-result” — this reframing helped the team evaluate AI as one tool among several rather than a mandate. I designed three fidelity levels: rule-based suggestions, hybrid (rules + ML ranking), and full generative search, with clear trade-offs for each.

The Decision & Why

I advocated for the hybrid approach because it balanced personalization gains with the predictability that a booking flow demands. Pure generative search introduced too much variability for a transaction-critical path — users need to trust that search results map to real, bookable inventory. The hybrid model let us use AI for intent understanding and ranking while keeping results grounded in structured data. This became the shipped approach and contributed to the team's Opus Research recognition as the industry's first Gen AI integration.

AI-First Customer Experience

Proposals get debated forever; working prototypes get adopted. Every major initiative — token system, Pixel Radar, MCP handoffs — started as something I built and demoed to stakeholders. This wasn't about going rogue; it was about reducing the abstraction barrier so cross-functional teams could evaluate ideas based on reality, not speculation.

In a merger environment with high turnover risk, I treated documentation as a first-class design artifact. Every system I built came with written rationale — not just “what” but “why this and not that.” The Liftoff mentorship program I initiated was partly about knowledge distribution: ensuring design thinking wasn't concentrated in any one person.

Pixel Radar v1 did exactly one thing. The token system started with one feature. The AI search launched as a hybrid, not full generative. In an organization with deep institutional skepticism toward “design,” the fastest path to influence was demonstrable, narrow wins that compounded into strategic credibility.

In Practice — Dev Handoff

Design systems only matter if engineering implements them accurately. Handoff at Air India was fragmented — Figma links in Slack threads, specs that didn't match builds, endless back-and-forth. I implemented a design-dev handoff workflow using Model Context Protocol — bridging design and engineering through AI-assisted tooling.

{ "component": "Button",
  "tokens": { "color": "--brand-red" }}

Team Recognition During My Tenure

Red Dot Design Award (displayed in Singapore Museum), Gold Stevie Award (first Indian airline), APEX Four Star (Most Improved Airline), World Travel Award for Asia's Leading IFE, and Opus Research recognition for industry's first generative AI integration. These are team achievements — I contributed to the design infrastructure and AI work that supported them.