Engineering the “Always-Fast” Product Surface At Scale

Ujjwal Gulecha, a seasoned tech expert, shares his journey from DoorDash to Anthropic, detailing how he elevated performance management from a technical debt to a core business lever. Discover his innovative strategies for scaling platforms, drastically reducing latency, and fostering a culture of reliability that drives growth and innovation.

In consumer technology, performance is rarely celebrated when it works. Users expect speed. They assume reliability. And when either falters, the impact is immediate, measurable, and unforgiving. For high-scale product surfaces that process tens of millions of requests, latency is not a backend metric. It is a growth constraint.

Ujjwal Gulecha, Member of Technical Staff and a technology expert at Anthropic working on the Claude Developer Platform, has spent nearly a decade operating at that intersection of performance, reliability, and product velocity. Prior to this, he spent six years building and scaling consumer platforms at DoorDash, progressing from Software Engineer to Engineering Manager while leading cross-functional teams across backend, iOS, Android, and web. His work focused on discovery systems, promotions infrastructure, and most notably, homepage performance management across platforms serving millions of users daily.

With experience spanning product engineering and platform strategy, and as a Judge at the Business Intelligence Group, Gulecha approaches performance not as a reactive firefighting function but as a product discipline in itself.

Performance as a Tier-0 Business Lever

One of the most consequential initiatives of his career was the multi-year Homepage Performance Management program from Q4 2022 through Q4 2024. The homepage represented roughly half of marketplace volume. It was the first surface users saw. And when Gulecha assumed end-to-end ownership, performance was far from acceptable. P90 latency stood at 5.5 seconds on iOS, 4.5 seconds on Android, and an alarming 24 seconds on the web.

For a Tier-0 surface, those numbers were not just technical debt. They were business risk.

As E2E owner and Tech Lead, Gulecha defined the long-term strategy and roadmap for a comprehensive latency transformation. He led a team of engineers, built a structured performance management framework, established SLO-style contracts with more than 20 dependency teams, and embedded latency as a first-class metric in cross-functional planning.

Running Performance Like a Product

The methodology was deliberately multi-dimensional.

First, observability. Previously, performance degradations could take hours or days to diagnose. Gulecha built systems and dashboards that reduced detection time to under 10 minutes, often before SLO burn became visible. That shift alone transformed operational responsiveness.

Second, structured prioritization. The team created a repository of all potential optimization levers, each mapped to effort and expected ROI. Instead of ad hoc tuning, performance improvements became data-driven investments.

Third, process discipline. Quarterly latency goals were introduced for dependency teams such as Promotions. SLO contracts formalized expectations and accountability across microservices, reducing uncontrolled fan-out and payload growth.

Technically, the improvements focused on what actually moved the needle across platforms. Reducing backend fan-out across 20 plus microservices. Shrinking payload sizes. Tightening endpoint efficiency. Improving viewport and hitch performance on Android. Ensuring reliability above 99.9 percent error-free sessions.

The results were transformational: iOS P90 latency reduced from 5.5 seconds to 1.4 seconds, a 75% reduction, Android reduced from 4.5 seconds to 1.2 seconds, a 73% reduction, Web reduced from 24 seconds to 3.5 seconds, an 85% reduction, error-free sessions maintained above 99.9% and degradation detection time reduced from hours or days to under 10 minutes.

The financial impact was equally significant. Android viewport optimization contributed to marketplace activity and new-vertical growth in the tens of millions of dollars. SLO-driven endpoint optimizations also delivered meaningful annual cost savings, while additional revenue benefits came from stronger conversion performance and lower abandonment rates.

The initiative became the first comprehensive multi-year performance management program for the homepage. Its payload tracking system was later adopted company-wide, and the SLO contract framework became the model for cross-team performance accountability.

Writing and Judging at the Systems Level

Gulecha’s thinking on performance extends beyond implementation. In his Hackernoon article titled "Why 'Obvious’ Performance Optimizations Often Backfire: Lessons From Systems Serving 50M+ Requests", he explores how superficial tuning can create downstream instability when systems operate at scale. The piece reflects lessons learned from real-world platform constraints, where local optimizations must be evaluated in the context of distributed dependencies.

His systems-level perspective also informs his role as a Judge at the Business Intelligence Group, where he evaluates innovation initiatives through the lens of durability, scalability, and measurable impact.

Across his career, he has also been recognized internally with an "Above Expectations" rating, specifically citing his end-to-end ownership of homepage performance management and ability to define both short-term and long-term strategies for high-priority technical problems.

The Broader Discipline of Reliability

What distinguishes Gulecha’s work is not just the magnitude of performance gains, but the reframing of performance as an organizational contract. In high-change environments where experimentation and feature velocity are constant, latency and reliability can easily degrade unless actively governed.

By treating performance guardrails, payload budgets, and regression detection as product requirements rather than engineering afterthoughts, he helped institutionalize a culture where speed and reliability coexisted with innovation.

Today, as he works on developer platforms powering advanced AI systems, the same principles apply. Whether in consumer marketplaces or AI tooling ecosystems, latency shapes trust. Reliability shapes retention. And experimentation only works when foundational surfaces remain fast and stable.

Through sustained execution, systems thinking, and cross-functional leadership, Ujjwal Gulecha has demonstrated that performance is not merely a technical metric. It is a strategic asset that directly influences user behavior, revenue, and long-term platform resilience.