Paperchase

Inspiration

I work at the MIT Media Lab as an ML researcher, and part of why I was hired was because I have a track record of building products quickly. I realized a lot of researchers are unfamiliar with how to validate startup ideas, and so I built Paperchase to solve that problem.

What it does

Takes an ArXiv paper and creates a VC analyst-style investment report identifying the viability and possible GTM strategy.

How I built it

Used Codename Goose to build out the MCP integrations, messed around with Apify actors until I found some good ones for Crunchbase and ArXiv.

Challenges I ran into

Generating insights that weren't a wall of text and simultaneously were qualitative enough to be useful was quite a task.

Accomplishments that I'm proud of

It works! And I ran it on some of my own papers, and it produced very impressive results.

What I learned

MCP servers being stateful versions of APIs seem a little pretentious when websockets exist. Or so I thought! As it turns out, the spec is very well-built when it comes to being used by agents themselves, and so my agentic system performs a lot better than I thought it would.

What's next for Paperchase

I'd like to turn this into an end-to-end system which is actually able to create an MVP version of the product born out of the research paper.

Built With

• apify
• codename-goose
• firecrawl
• next
• perplexity