Event-driven document intelligence
AI-Powered Document Classification
Processed incoming business documents asynchronously without blocking the product experience.
- Role
- Backend architecture and AI pipeline development
- Focus
- AI systems and product infrastructure
- Stack
- Python · Next.js · Redis · Dramatiq
The engineering story
01 · Constraint
The challenge
Documents arrived in different formats and volumes, while OCR quality varied enough that a single extraction path could not be trusted.
02 · System
The approach
Separated webhook ingestion from processing with Redis and Dramatiq, then built format-aware extraction pipelines with an OpenAI Vision fallback for complex documents.
03 · Delivery
What shipped
- Handled PDF, DOCX, and XLSX inputs
- Kept request paths responsive with background processing
- Added model fallback for difficult extraction cases