Women's health evidence exists, but it is structurally hard to reach. High-quality clinical research is fragmented across specialist medical databases, each with its own access logic, terminology, and institutional barriers. These systems were built for academic continuity, not for clinicians under time pressure, advocates building policy arguments, or practitioners working outside universities. The gap was not credibility. It was discoverability. What was missing was a neutral discovery layer that could surface medical evidence across systems while preserving methodological standards and removing unnecessary institutional friction. After launch, feedback revealed a second challenge: the original blue and pink palette, intended to be approachable, read as gendered to some users and risked undermining the platform's authority in medical research.

I built a research discovery interface that fans a single query out across four default sources in parallel, PubMed, EuropePMC, OpenAlex, and ClinicalTrials.gov, with thirteen further opt-in sources available behind advanced controls when a user wants broader coverage: bioRxiv, medRxiv, SciELO, WHO publications, CORE, OpenAIRE, Semantic Scholar, and six traditional medicine corpora. The primary search path routes through a Supabase edge function that owns the rate-limit budget for the whole platform, so user volume does not translate into per-user upstream traffic and the service stays free at any scale. The first product shape came from Figma and Figma Make: fast enough to test structure, language, and the research workflow before over-investing in code. Once the surface needed real interaction quality, database connectivity, and maintainable architecture, I rebuilt it in Claude Code with React, Tailwind v4, and shadcn/ui on top of Radix primitives. Dialogs, dropdowns, tooltips, tabs, and selects are handled by accessible component foundations; styling is driven through a token system rather than one-off visual decisions. I still owned the design work Figma Make does not solve: layout, responsive states, hierarchy, density, and the middle layer between a promising prototype and a product people can actually use. On top of the search layer, a custom extraction pipeline surfaces sex-equity signals from each abstract using pattern matching, not an LLM, and labels every signal with where it came from (structured API field, full text, or abstract). When Semantic Scholar is enabled, an optional one-sentence TLDR sits beside the result so researchers can triage faster without leaving the list. Two search modes are designed in: rapid clinical discovery and deep research exploration, both with advanced filtering by study type, population, and evidence strength. Following that feedback, I ran a major redesign, pivoting from the gendered blue and pink palette to a deep burgundy editorial system, keeping pink only as a restrained accent. The decision came from user feedback that the original colours read as "femtech marketing" rather than clinical trust, and from the strategic need to build credibility while not reinforcing gender bias in healthcare design.