Developer docs platform with instant search and versioning.

Helix sells APIs to regulated teams. Their docs lived in a static generator that could not express versioning cleanly, search returned stale snippets, and internal writers avoided publishing because the build was fragile. Sales engineers became the human search index — not scalable.

The VP Engineering asked for a credible migration before a flagship conference — six weeks for parity, four for polish. No downtime on public URLs. Maria led engineering; we paired with their developer advocacy lead from day two.

We rejected “move everything to a wiki” pressure from an internal tools team. The constraint was auth-aware docs with the same SSO model as the product console.

// TODO: Verify with client legal before publishing — product positioning and timelines.

Framing. Documentation is a release artefact, not a marketing pamphlet. That meant CI gates, preview environments per PR, and owners on every nav section.

Architecture. Next.js + MDX with remark plugins for link checking and code sample extraction. Algolia indices split by release channel; middleware routed enterprise tenants to the correct index without leaking prerelease content.

Phasing. Week 1–2: inventory + redirect map. Week 3–4: MDX migration with automated diff reports for SMEs. Week 5–6: cutover behind a flag; week 7–10: search tuning and snippet design.

Judgment. We insisted on deleting duplicate pages instead of hiding them — politically spicy, SEO-correct for developer properties.

// TODO: Verify with client legal before publishing — search and auth integration details.

Authenticated docs weekly active readers rose 62% after launch, while median time-to-first-call on API onboarding tickets dropped 34% using Helix’s support taxonomy. Migration landed in the committed six-week window; search cluster maintained 99.2% uptime during the conference spike.

Developer advocates resumed publishing because the pipeline felt like product engineering. Helix retained us lightly for index tuning; the programme closed with a contributor guide their open-source community adopted.

// TODO: Verify with client legal before publishing — metrics definitions and uptime claims.

Writers hated the first lint rule set — too noisy. We dialled rules back and shipped warnings before errors, which improved adoption. We underestimated how often legal needed redacted examples; we added a fenced “synthetic payload” pattern mid-programme.

Next time we would prototype search snippets with real customer questions in week one, not week seven. We also learned to schedule content freeze windows earlier around conferences.

// TODO: Verify with client legal before publishing — internal adoption dynamics.