Prompt Engineering Workbench (TypeScript, React, PostgreSQL, LLM APIs, Redis) overview

A developer-facing prompt engineering workbench that provides a React-based playground for authoring, versioning, and A/B testing prompts against multiple LLM providers. PostgreSQL stores prompt versions, evaluation datasets, and scored results. Redis caches LLM responses for rapid iteration during prompt development. The TypeScript backend (via Hono) proxies LLM API calls with token tracking, cost attribution, and latency measurement. Zod validates prompt templates and evaluation schemas. Designed for AI product teams and prompt engineers who need structured experimentation beyond ad-hoc notebook workflows. The tradeoff is evaluation subjectivity — automated scoring captures surface-level quality but often requires human rating for nuanced prompt quality assessment.