Build an evidence database that rates supplement claims against studies

A nutrition scientist with no developer background used Claude Code to build a bilingual site with 504 compound profiles, 1,794 rated claims and about 7,000 cited sources, with agents checking each other's work.

Done withClaude Code· Nov 15, 2026

What they did

A Redditor who works as a nutrition scientist, and says they are not a developer, spent a few weeks building synaptiklabs.org with Claude Code. The site covers supplements, nootropics, peptides and medicines in English and German. Multi-agent workflows draft each compound profile from the primary literature, then separate agents try to refute the draft. Every DOI and PMID is checked automatically against Crossref and PubMed, so an invented citation fails. Each mechanism figure is an SVG drawn by one agent and reviewed by two others, one for the science and one for design and accessibility. Claude Code also wrote the Next.js components and the test suites. For overnight runs the poster sets a goal and a budget, and Claude runs agents in parallel git worktrees, keeps a ledger and only asks about real decisions.

How it went

The site is live with 504 profiles, 1,794 claims each given an evidence level from 0 to 4 and a separate result, about 7,000 sources and 47 interactive figures. Claude wrote about 1,600 unit tests and roughly 2,000 Playwright browser tests, and nothing merges without green tests. The poster still sets the rating rules and checks ratings and key statements by hand. Parallel agents ran the laptop out of memory more than once. In the comments the poster notes the site finds nothing new: of 25 popular claims in its hype check, 7 held up as advertised.

Worth knowing

The poster credits the gains to checks scripted around Claude, such as resolvable citations and validation scripts, rather than trusting the generated text. Parallel agents need hard limits on how many run at once. Gene-by-supplement advice was left out on purpose because it drifts into personal medical advice.

Try it

With Claude Code· 5 parts to fill in

Help me build a reference on [topic] where every claim is checked against primary sources. For each [item], draft a profile from the literature, then run a separate pass whose only job is to refute the draft. Resolve every citation against [Crossref, PubMed or another index] and fail the item if any reference does not resolve. Rate each claim using [my rating rules]. Write tests for every page and do not merge anything until they pass. Run at most [number] agents at once, keep a ledger of what each did, and stop to ask me before changing the rating rules or publishing.

Square brackets are the parts to fill in. The prompt is ours; the story is theirs.

Where they said it

Related