GenAI Data Extraction — Aryan Iyappan
PDF → structured data, end-to-end on AWS
Velocius AI Labs · Aug – Nov 2024
PDF → structured data, end-to-end on AWS
Developed a robust solution for transforming unstructured PDF documents into structured, actionable data, from idea to a working prototype. Leveraging Python, LangChain, FastAPI, and React, with a highly scalable architecture on AWS — Lambda, API Gateway, DynamoDB, VPC for secure processing, and S3 for storage. The React SPA is hosted on CloudFront via S3, with CI/CD powered by CodeBuild and CodePipeline.
Structured JSON Before JSON Mode
Built robust extraction pipelines when LLMs couldn't reliably output structured JSON — no native JSON mode existed in APIs yet. Engineered parsing, validation, and retry strategies that turned unreliable model outputs into deterministic data.
Tags: JSON MODE, RETRY LOGIC
End-to-End AWS Architecture
Lambda, API Gateway, DynamoDB, VPC, S3, CloudFront. Serverless-first, horizontally scalable, VPC-isolated for secure document processing. Each piece purpose-built, not over-provisioned.
Tags: LAMBDA, VPC, S3
Full-Stack Delivery
Python + LangChain + FastAPI backend. React SPA frontend. CI/CD via CodeBuild and CodePipeline. From idea to working prototype — not a proof of concept, a production system.
Tags: FASTAPI, REACT, CI/CD