Systems & AI
AI-Powered Document Analysis System
End-to-end RAG system that turns PDFs into grounded answers.
- When
- Aug 2025 – Dec 2025
- Headline result
- 85% accuracy · 5,000+ queries
- Architected an end-to-end RAG system with React, Node.js, and GPT-4 to transform PDFs into grounded answers, achieving 85% accuracy across 5,000+ evaluated queries.
- Built a semantic retrieval pipeline that segmented PDFs into 512-token chunks and indexed OpenAI embeddings in Pinecone, supporting 1,000+ daily queries.
- Reduced document-processing time from 45 seconds to 8 seconds (82%) by decoupling ingestion with SQS and Lambda and parallelizing chunking and embedding generation.
Tech stack
- React
- Node.js
- GPT-4
- OpenAI API
- Pinecone
- AWS Lambda
- AWS SQS