Stock-Exchange Announcements Feed on AWS – Parallel Serverless PDF Parsing

Project Overview

Built an event-driven AWS pipeline that ingests announcement PDFs and metadata files from an exchange FTP feed (synced to S3 every minute over an IPsec VPN) and parses each document in parallel: a splitter groups pages into chunks, Lambda workers extract them with one reused converter, and a merger writes data, tables and summary JSON. Idempotent DynamoDB claims prevent reprocessing, S3 chunk manifests removed the page limit on large PDFs, and Google Vision OCR with a combined per-document JSON completes delivery.

Tools & Technologies: AWS Lambda · Step Functions · S3 · EventBridge · DynamoDB · ECR · Docker · Python · Docling · Google Cloud Vision OCR · IPsec VPN · FTP