/morph

Over 50% of research papers are never cited. Our system maps ideas across domains and systematically explores new combinations, helping researchers uncover insights they would never find on their own.

Comment

Inspiration

Over 50% of research papers never get cited. Not because they're wrong — but because no one ever connected them to the right problem. Breakthroughs don't come from thin air. They come from combining existing ideas: take concept A, combine it with concept B, and you arrive at something new — concept C. Humans can only hold a few ideas at a time. AI can explore thousands of combinations simultaneously.

What it does

/morph is an AI-powered discovery engine that systematically combines ideas across research domains to surface non-obvious connections. Instead of searching for what's already known, /morph asks: what can be discovered by combining what's known?

Give it a research problem, and it maps concepts across fields — biology, materials science, chemistry, engineering — finding structured, explainable links that a human researcher would never realistically stumble on. Not guesses. Not summaries. New combinations worth exploring.

How we built it

  • Frontend: Next.js, React, and TypeScript, hosted on Vercel
  • AI layer: OpenAI and Claude working under the hood to parse, embed, and reason across research concepts
  • Backend: A self-hosted instance of OpenClaw running on a VPS deployed via Fly.io, handling the combinatorial logic and knowledge mapping; see https://fly.io/docs/blueprints/deploy-openclaw

Challenges we ran into

The biggest challenge is signal vs. noise. If you have 1,000 concepts, the number of possible combinations explodes into the billions — and 99.999% of them are meaningless. Building a system that ranks meaningful combinations over noise was the core technical hurdle.

Accomplishments that we're proud of

We built a working system that can take a research problem and surface genuinely surprising cross-domain connections. The first time it linked a concept from materials science to a biological mechanism we hadn't considered — and the link actually made sense — that was the moment it clicked.

What we learned

Idea generation is the easy part. The hard part is making combinations meaningful and actionable. We also learned that the real bottleneck in research isn't access to information — it's the inability to see across fields. Researchers are deep in their domains but rarely have time to explore adjacent ones.

What's next for /morph

  • Deeper domain-specific models starting with biotech and drug discovery
  • Letting users validate and rate combinations to improve ranking over time
  • Building a shareable "discovery graph" so teams can collaborate on combinatorial exploration
  • Partnering with research institutions to test /morph on real unsolved problems

Built With

Share this project:

Updates