Synopsys

01Problem

Stakeholders needed one product surface to evaluate several LLMs against the same criteria and to run trial inference — instead of disconnected notebooks and one-off demos — before choosing a model for NLP, support automation, content generation, or analysis.

02Selected contributions

Built and maintained the reusable React / Next.js UI for Synopsys’s internal LLM evaluation product: comparison views, trial-inference flows, and the component layer those screens share. Sequenced frontend work by effort versus impact, and used code review and implementation practice to keep TypeScript and UI conventions consistent across the team. Worked with product and design so UX and backlog order matched what stakeholders actually needed to decide.

• Reusable UI for model comparison and trial-inference flows
• Component and styling conventions on Next.js / Tailwind CSS
• Backlog sequencing that weighed engineering cost against decision value
• Frontend standards shared through implementation and review

03System context

React and Next.js application UI, typed in TypeScript and styled with Tailwind CSS, talks to trial-inference and comparison backends owned elsewhere. Docker is part of the delivery environment. Training pipelines and model weights were out of scope.

04Scale & constraints

• Frontend ownership of the evaluation product, not model training or weights
• Reusable components had to match an existing design language
• Delivery sat inside an established product team and Docker-based environment

05Outcome

The evaluation product gained reusable frontend surfaces instead of one-off screens, and the team inherited shared UI conventions. Model quality, latency, and selection metrics are not published.

• Reusable UI for an enterprise LLM evaluation product
• Shared frontend standards on the delivery path

06Stack

• React
• Next.js
• JavaScript
• TypeScript
• Tailwind CSS
• Docker