-
-
PAPER PILOT ARCHITECTURE
-
FRONT PAGE ARCHITECTURE
Inspiration
Born from the experience and needs of researchers, this project is designed to advance research itself. We, the builders, are current research students who have had reasons to question our life choices due to the vast amount of research resources available, but we are unsure where to begin and how to utilise the information at our disposal. Flipping through hundreds of research papers to find the most suitable one for an area of research is a major problem in academia, and we hope to solve it with Paper Pilot.
What it does
Paper Pilot serves as an AI agent that takes in input from a user, even as little as 'one word', and produces a comprehensive list of every research paper where that word has been mentioned. It then goes further to analyse those papers and produces a well-structured literature review that the user can use for their own research, thus reducing the hours and days that would have been spent by the user doing this manually.
How we built it
The project was built using NextJS - Frontend, API - Gateway, Lambda - API(Proxy), Bedrock agent orchestrator + multiple Lambdas for tools(searchArxiv, fetchPdf, parsePdf, formatCitations), S3 for PDFs and extracted texts, DynamoDB for metadata/cache and CloudWatch for logs/metrics.
Challenges we ran into
Challenges we ran into includes; picking the right model for the ON_DEMAND inference type, integrating the tools to the agentcore.
Accomplishments that we're proud of
The main accomplishment here is being able to build an AI agent that reduces the amount of time researchers would ordinarily spend discovering and reviewing suitable literature to aid their research venture.
What's next for PaperPilot
The next step for PaperPilot is making it available for researchers to use and advance the academic world.
Log in or sign up for Devpost to join the conversation.