Elyvite
An AI research ecosystem for health professionals: from raw source to published article, all in one system.
- Role
- Founder, AI product designer, full-stack development
- Period
- September 2025 to today
- Team
- Just me, plus an AI agent team I built for it
- Status
- Open beta since June 2026
There are plenty of good sources. Just not in one place.
Health knowledge is scattered across podcasts, videos, studies and databases. Anyone who wants to research properly spends hours comparing sources and still ends up facing contradictory expert opinions. Elyvite shows where experts agree and where they do not, with sources for every statement.
Elyvite Pipeline
macOS app. The data machine that turns raw material into searchable knowledge.
Elyvite Platform
The workspace where users establish and substantiate expert consensus.
Elyvite Website
Multilingual marketing website with its own AI sales assistant.
Making consensus visible, contradictions included.
Every statement carries its source. The answer shows agreement in percent, right next to the points where experts disagree. Video sources open directly and can be saved into categories. On top of that, a delta check between expert opinion, the state of research and web sources.
Trust is calibrated, not faked.
Expert knowledge becomes daily content.
The same knowledge base produces daily news and studies, expert briefings on individual specialists, and longer feature articles per category. Three pipelines run automatically; approval stays with me. A product that needs fresh content every day cannot afford manual labour.
- 01
Collect sources
Studies, preprints, trade news and expert talks arrive daily and are sorted by category.
- 02
Split briefings
Expert statements become briefings, cut into sections and tagged by topic.
- 03
Build articles
Per category, an agent pulls the matching sections together and writes a feature article with a source list.
- 04
Review and approve
No article is published without review.
- 05
Publish
On approval, the article goes live in six languages.
Elyvite Pipeline.
Sources come in, get translated, processed and indexed. Every run is traceable: transcript, translation, preparation, database, memory. Everything runs locally on the Mac.
Not a prototype. A running product.
Three products that run on their own and form one system together. Plus user management with mandatory 2FA, subscriptions and credits, analytics, billing via Lemon Squeezy, and three automated editorial pipelines.
It did not stay a prototype. Built single-handedly, backed by a structured multi-agent workflow: product strategy, UX, retrieval architecture, full-stack development and operations mesh together instead of running side by side.
Sensitive health context, traceability of every source, GDPR and EU AI Act transparency duties, prompt injection defenses, fluctuating source and model quality, a small budget.
Figma · Tauri · Rust · Next.js · React · TypeScript · Python · FastAPI · Pydantic AI · Supabase · PostgreSQL · pgvector · SQLite · Zep · OpenRouter · Sanity · Lemon Squeezy · Docker · Vercel · Render · Playwright · Vitest · pytest
From product strategy through UX and agent orchestration to operations and security, everything here is connected and runs in production.
Sounds like your project? Say hello. 30 minutes, and you will know whether it fits.