Linky

Hackathon prototype that turns a URL into a searchable knowledge base with retrieval-backed answers.

Hackathon Winning ProjectPublished January 28, 2024Updated March 11, 2024

  • TypeScript
  • AI
  • RAG
  • PostgreSQL
  • Mantine
Linky hero image

Context

Linky started with a small but useful prompt: what if one URL could become a structured place to learn, not just a one-off chat input.

I wanted the project to show both retrieval quality and a calmer product experience, since many AI demos only prove the model call and not the interface around it.

Build

The product flow stayed intentionally simple: start with a URL, ingest the source material, and turn it into a knowledge layer that supports exploration instead of generic answers.

Under the hood, the stack paired retrieval, storage, and a typed frontend so the project felt like a usable tool with real product decisions behind it.

Takeaways

The prototype showed how ingestion, retrieval, and interface design can reinforce each other when the product starts with a clear learning workflow.

It also became a stronger write-up because the product decisions and technical implementation could be presented with the same level of clarity.