All workLive on AWS

Research Paper Summarizer

A serverless AWS pipeline that summarizes an arXiv paper in about 30 seconds, and costs nothing while idle.

NYU Cloud ComputingLive on AWSCDKStep FunctionsBedrock2026LiveDemo filmGitHub
Research Paper Summarizer landing page
Live site
Research Paper Summarizer demo film cover
Demo film
RequestAPI GatewayQuotareserve, refundDedupSHA-256 in DynamoDBMap-reduceStep Functions, 5 at a timeSummary, chat, graphBedrockSeen before: cached result returned without a second run

What it does

Search arXiv, pick a paper, and get a structured summary, a chat that cites the paper, and a knowledge graph of its concepts. I was the sole architect and primary engineer.

Problems it solves

  • Lambda stops at 15 minutes, so a Step Functions map-reduce fans the paper out, capped at five concurrent chunks to respect Bedrock's rate limits.
  • Step Functions caps state at 256 KB, so stages pass S3 keys rather than text.
  • Two people summarizing the same paper pay once: a SHA-256 content hash in a DynamoDB index returns the cached result.
  • Queue redeliveries cannot run a job twice, because the execution name is the job ID.
  • A quota that reserves before it spends and refunds on failure, so a burst of requests cannot overrun the budget.
  • Different models for different jobs on Bedrock: a large open model for the high-volume summaries, Claude for chat and graph extraction.

Next